COMMON LOON (GAVIA IMMER) BIOGEOGRAPHY AND REPRODUCTIVE
SUCCESS IN AN ERA OF CLIMATE CHANGE
By
Allison Byrd
B.A. University of Rhode Island, 2001
A THESIS
Submitted in Partial Fulfillment of the
Requirements for the Degree of
Master of Science
(in Ecology and Environmental Science)
The Graduate School
The University of Maine
May, 2013
Advisory Committee:
Brian Olsen, Assistant Professor of Biology & Ecology, Advisor
Rebecca Holberton, Professor of Biology
David Evers, Executive Director, Biodiversity Research Institute
ii
THESIS ACCEPTANCE STATEMENT
On behalf of the Graduate Committee for Allison Byrd I affirm that this
manuscript is the final and accepted thesis. Signatures of all committee members are on
file with the Graduate School at the University of Maine, 42 Stodder Hall, Orono, Maine.
Brian Olsen, Assistant Professor of Biology & Ecology December 3, 2012
LIBRARY RIGHTS STATEMENT
In presenting this thesis in partial fulfillment of the requirements for an advanced
degree at The University of Maine, I agree that the Library shall make it freely available
for inspection. I further agree that permission for “fair use” copying of this thesis for
scholarly purposes may be granted by the Librarian. It is understood that any copying or
publication of this thesis for financial gain shall not be allowed without my written
permission.
Signature:
Date:
COMMON LOON (GAVIA IMMER) BIOGEOGRAPHY AND REPRODUCTIVE
SUCCESS IN AN ERA OF CLIMATE CHANGE
By: Allison Jeanne Byrd
Thesis Advisor: Dr. Brian J. Olsen
An Abstract of the Thesis Presented
in Partial Fulfillment of the Requirements for the
Degree of Master of Science
(in Ecology and Environmental Science)
May, 2012
Climate change has the potential to shift and restrict ranges for a suite of species.
The birds of the boreal ecosystem, like the Common Loon (Gavia immer), may be
particularly at risk given the changes predicted for this biome. Current range models for
this iconic water bird predict that large sections of the United States may lose the loon in
the next 100 years, but these models are based on habitat correlations and not the
demographic mechanisms that will actually produce the change. The primary goal of our
research was to understand the factors that determine loon vulnerability to climatic
change at multiple scales. We applied a recursive partitioning technique to analyze loon
presence/absence in 288 lakes across the southern edge of their North American
distribution using 112 abiotic and landscape-level factors. The resulting binary tree
(“decision tree”) classified lakes into groups based on the probability of loon presence,
while maximizing homogeneity within the resultant two nodes. The most significant
splits in the cross-validated tree were created using using lake salinity, acidity, and
sulfate levels. We employed similar methods to compare loon occupancy and seasonal
fecundity at a smaller scale (New England) to elucidate potential demographic
mechanisms of loon persistence. Results from twenty potential predictors suggest that
similar processes are driving loon presence/absence both continentally and within New
England (lake salinity, alkalinity, and lake surface area). Loon productivity, on the other
hand, was best predicted using the size of the lake and its drainage basin. Lake surface
area and characteristics of the drainage basin are thus good predictors of both loon
distribution and loon productivity, and are thus also likely to be useful in predicting range
shifts in the future. As few (if any) of the predictors of productivity in the best decision
trees are likely to change dramatically with climate, these outcomes suggest that future
range alteration for loons due to climate change are likely to be more sensitive to annual
adult survival (which will influence breeding ground settlement patterns) than
environmental factors encountered on the breeding grounds.
After highlighting the environmental factors that predict loon occupancy and
productivity, we explored environmental predictors of individual energetic condition.
Physiological measures may offer more information about the likelihood of loon
persistence, because they can identify covariance between energetic condition and
breeding habitat quality. We used blood metabolites and behavioral observations to
evaluate the energetic costs of common loons (Gavia immer) breeding on lakes across a
gradient of environmental, spatial, and social conditions. Using samples collected over
two years (n=97) along the species’ southern breeding range edge (ME, NH, MA, MT,
WA), we identified the number of offspring, daily maximum temperature, and latitude as
the most important drivers of the energetic cost of maintaining a breeding territory.
Specifically we found that: 1) loons with two chicks expend more energy than those with
one, 2) loons near the southern range edge expend more energy to produce a given brood
size than those nearer the range center and, 3) birds breeding in warmer temperatures
expend more energy than those in cooler temperatures (controlling for year, territory
type, and calendar date, free glycerol levels, size-corrected body mass, and longitude).
We suggest that as environmental conditions change in the coming years, blood
metabolites offer a promising predictor of population collapse along range boundaries.
Energetic condition deteriorated toward the southern range edge and in warmer
conditions, controlling for the number of offspring produced, which suggests that loons
may be sensitive to increasing global temperatures. We suggest that as environmental
conditions change in the coming years, blood metabolites offer a promising predictor of
population collapse along range boundaries.
iii
ACKNOWLEDGEMENTS
I would like to thank the numerous individuals who made this thesis possible. I
extend a sincere thank you to my committee members David Evers and Rebecca
Holberton for their support. I would also like to thank William Halteman for his
statistical expertise and encouragement as I made my way through the world of data
analysis. I have countless peers who made project possible including Kate Ruskin, Ellen
Robertson, Dave Grunzel, Evan Adams, Leah Culp, Chris Tonra, Adrienne Leppold,
James Style, Anna Henry, Laura Kennedy, Anya Rose, Matt Jones, and also Kate Taylor,
Carrie Osborne, Carl Anderson, Jeff Fair, Alison Jackson, Ginger and Dan Poleschook,
Kristin Ditzler Strock, and Courtney Wigdahl. I cannot sufficiently thank Maureen
Correll, Daniel McConville, and Jenny McCabe for their friendship and support during
my time at UMaine. I’d like to thank Wren for countless laughs, runs, and canine
support. I would like to thank Sue Anderson and Trish Costello for their administrative
assistance throughout my time at the University of Maine. My family provided
encouragement and love, including Barbara, Ray, Adam Byrd, and Melanie, Dan, Lilly,
and Morgan Kerr. Funding was provided by the School for Biology and Ecology,
Biodiversity Research Institute, the Maine Agriculture and Forestry Experimental
Station, and the Northeast Temperate Network of the National Park Service. Field
assistance was provided by Mike Chickering, Tori Polito, Keith Blanchette, Matt O’Neal,
Ashley Malinowski, Seth Wile, and Misty Libby. I’d like to thank David Bishop for
believing in me and putting me in contact with Brian Olsen, which is how this incredible
journey all began
iv
Finally, I would like to extend the most heartfelt thank you possible to my
advisor, Brian Olsen. Brian, you have showed me what it means to balance brilliance,
encouragement, humor, editing, teaching, motivation, family, exercise, grading,
enthusiasm, sarcasm, a seven-plus member lab, confidence, humility, and constant
learning… all without losing your mind. I cannot adequately express how much I have
learned from you during my time here and I could not be more proud to be counted as
one of your first students. You are an incredible teacher and mentor, and I will be forever
thankful for your guidance. Thank you for always believing in me. I dedicate this thesis
to you.
vi
TABLE OF CONTENTS
ACKNOWLEDGEMENTS ............................................................................................................. 1
LIST OF TABLES ........................................................................................................................... 1
LIST OF FIGURES ......................................................................................................................... 1
Chapter
1. COMMON LOON (GAVIA IMMER) BIOGEOGRAPHY AND REPRODUCTIVE
SUCCESS IN AN ERA OF CLIMATE CHANGE: BACKGROUND
AND CONCLUSIONS ............................................................................................................... 1
1.1. DISTRIBUTION ...................................................................................................... 1
1.2. HABITAT ................................................................................................................. 2
1.3. FEEDING ................................................................................................................. 3
1.4. PHYSICAL CHARACTERISTICS ......................................................................... 3
1.5. BEHAVIOR.............................................................................................................. 4
1.6. BREEDING INFORMATION ................................................................................. 5
1.7. SITE FIDELITY ....................................................................................................... 6
1.8. MANAGEMENT AND CONSERVATION ............................................................ 8
1.9. RANGE EDGES ....................................................................................................... 8
1.10. CLIMATE CHANGE ............................................................................................. 9
1.11. MODELING SPECIES DISTRIBUTION IN RESPONSE TO
CLIMATE CHANGE .......................................................................................... 10
vi
1.12. LAKE METRICS PREVIOUSLY SHOWN TO AFFECT LOON
PRESENCE/ABSENCE AND PRODUCTIVITY .............................................. 11
1.12.1. Lake Depth ........................................................................................... 11
1.12.2. Lake Surface Area ................................................................................ 12
1.12.3. Water Clarity/Turbidity ........................................................................ 13
1.12.4. Surface temperature .............................................................................. 13
1.12.5. pH ......................................................................................................... 14
1.12.6. Conductivity ......................................................................................... 15
1.12.7. Dissolved Organic Carbon (DOC)........................................................ 16
1.13. USING TRIGLYCERIDES AND FREE GLYCEROL TO DETERMINE
ENERGETIC COST ............................................................................................ 17
1.14. SYNOPSIS .......................................................................................................... 25
2. MACHINE LEARNING TECHNIQUES: A TOOL FOR UNDERSTANDING COMMON
LOON (GAVIA IMMER) BIOGEOGRAPHY AND PRODUCTIVITY IN AN ERA OF
CLIMATE CHANGE ............................................................................................................... 29
2.1. ABSTRACT ........................................................................................................... 29
2.2. INTRODUCTION .................................................................................................. 30
2.2.1. Climate Change and Freshwater Communities ...................................... 30
2.2.2. Machine Learning: Using Decision Trees and Random Forests to
Determine Variable Importance ............................................................ 32
2.2.3. General Approach ................................................................................... 34
2.3 METHODS .............................................................................................................. 35
2.3.1. Methods Overview ................................................................................. 35
vi
2.3.2. Continental-scale Range Descriptors ...................................................... 36
2.3.3. Regional-scale Range Descriptors .......................................................... 45
2.3.4. Regional Productivity Descriptors .......................................................... 46
2.4. RESULTS ............................................................................................................... 48
2.4.1. Continental-scale Range Descriptors ...................................................... 48
2.4.2. Regional-scale Range Descriptors .......................................................... 52
2.4.3. Regional Productivity Descriptors .......................................................... 55
2.5. DISCUSSION ......................................................................................................... 58
2.5.1. Continental-scale Range Descriptors ..................................................... 58
2.5.2. Regional-scale Range Descriptors .......................................................... 62
2.5.3. Regional-scale Productivity Descriptors ................................................ 63
2.5.4. Validating Envelope Models with Hierarchical Decision Trees ............ 64
2.5.5. Implications for Climate Change on the Range of the Common Loon .. 66
3. BLOOD METABOLITES AS INDICATORS OF BREEDING HABITAT QUALITY: A
SENSITIVE METHOD TO PREDICT RANGE ALTERATION UNDER CLIMATE
CHANGE? ................................................................................................................................ 68
3.1 ABSTRACT ............................................................................................................ 68
3.2 INTRODUCTION ................................................................................................... 69
3.3 MATERIALS AND METHODS ............................................................................ 73
3.3.1. Study population, field methods, and lab assays .................................... 73
3.3.2. Behavioral Observations ......................................................................... 74
3.3.3. Statistics .................................................................................................. 75
3.4 RESULTS ................................................................................................................ 78
vii
3.4.1. Establishing Baseline Metabolic Measures ............................................ 78
3.4.2. Triglyceride Levels In Breeding Loons .................................................. 79
3.4.3. Behavior ................................................................................................. 82
3.5 DISCUSSION .......................................................................................................... 83
3.6 SUMMARY............................................................................................................. 87
BIBLIOGRAPHY .......................................................................................................................... 89
BIOGRAPHY OF THE AUTHOR .............................................................................................. 101
viii
LIST OF TABLES
Table 1. Descriptive Statistics for Metabolite Values…………………………..…..…18
Table 2. Data Sources………………………………………………………….……......37
Table 3. Variables in National Presense/Absence Analysis……………….…..….…...40
Table 4. Variables in New England Presence/Absence Analysis…………….….…….46
Table 5. Variables in New England Productivity Analysis………………….….....…..48
Table 6. Descriptive Statistics for Triglyceride and Free Glycerol Levels………..….79
Table 7. Triglyceride Probability Models………………………………………..….....81
Table 8. Triglyceride Model-Averaged Parameter Estimates……………………...….82
Table 9. Triglyceride Levels of Loons That Escalated Interactions with Intruders…...83
ix
LIST OF FIGURES
Figure 1. Portion Of A Figure From Jenn-Eiermann and Jenni (1994) ...................................... 20
Figure 2. Figure From Jenni-Eiermann And Jenni (1992) .......................................................... 22
Figure 3. Figure From Jenni and Schwilch (2001) ..................................................................... 24
Figure 4. Determining Loon Presence/Absence ........................................................................... 37
Figure 5. Lakes Used In Analysis. ................................................................................................ 38
Figure 6. Map For Determining Lake Inclusion. .......................................................................... 39
Figure 7. United States Classification Tree .................................................................................. 49
Figure 8. Random Forest Analysis For United States ................................................................... 51
Figure 9. Dependence Plots for United States Analysis ............................................................... 52
Figure 10. Classification Tree For New England ......................................................................... 53
Figure 11. Random Forest For New England Analysis ................................................................. 54
Figure 12. Dependence Plots For New England Analysis ............................................................. 55
Figure 13. Regression Tree For Productivity Model .................................................................... 56
Figure 14. Random Forest For Productivity Analysis .................................................................. 57
Figure 15. Temperature/Year Interaction Plot. ............................................................................. 87
1
CHAPTER 1
COMMON LOON (GAVIA IMMER) BIOGEOGRAPHY AND REPRODUCTIVE
SUCCESS IN AN ERA OF CLIMATE CHANGE:
BACKGROUND AND CONCLUSIONS
STUDY SPECIES: THE COMMON LOON
1.1. DISTRIBUTION
The world population size of the common loon is estimated at 615,000 individuals
(Evers et al. 2010). Loons breed on freshwater lakes throughout Canada and several
northern states in the U.S. (including parts of Washington, Idaho, Montana, Wyoming,
North Dakota, Minnesota, Wisconsin, Michigan, New York, Maine, Vermont, New
Hampshire, and Massachusetts). The northern edge of their breeding range extends to
Alaska and the taiga shield forest ecosystem of Canada (Evers et al. 2010).
Historically, loons have been reported to breed in locations further south than
their current distribution with confirmed breeding in areas of Northeast California,
Northern Iowa, Southern Minnesota, Northern Illinois, Southern Wisconsin, Northern
Indiana, Southern Michigan, Northern Ohio, and Northeast Pennsylvania; however,
availability of information on past distributions is geographically and temporally
sporadic. Loons began breeding in Northern Massachusetts approximately 40 years ago
and in parts of Pennsylvania 30 years ago, which represents either a range expansion or a
recolonization of their former range in these areas (Evers et al. 2010).
2
Loons winter on the Atlantic and Pacific Oceans. In the Atlantic, the range
extends from northern Canada along the coast of Newfoundland to the Gulf of Mexico.
Pacific Ocean wintering distribution ranges from the Aleutian Islands off of Alaska to the
Gulf of California (Evers et al. 2010). Loons largely remain on inshore ocean waters and
will use inlets and coves during the winter. Loons prefer water depths less than 19 m
deep, but will also utilize waters 100 m deep and 100 m from shore, largely depending on
prey movement and availability (Kenow et al. 2009). While rare, there have been reports
of overwintering loons on freshwater systems in the southern United States (Evers et al.
2010).
1.2. HABITAT
During the breeding season, loons prefer freshwater lakes that are greater than 24
acres in size and that have irregular shorelines that create inlets and coves suitability for
nesting. Rivers are rarely used for nesting habitat unless large areas with little current are
available (Evers et al. 2010). Water quality characteristics are important for habitat as
loons are visual, underwater predators (Barr 1973). Adults feeding in turbid water spend
twice as long capturing prey as those in clear waters (Gostomski and Evers 1998). Total
lake surface area, depth, and surface temperature have been shown to be significant
predictors of loon presence or absence on lakes in New Hampshire (Blair 1992).
Loons also use lakes and large rivers as staging areas during spring and fall
migration to rest and forage. These lakes and rivers have similar characteristics to
breeding lakes, such as clear water and abundant prey (McIntyre and Barr 1983).
Typically young will leave their natal territories and not return to freshwater lakes until
3
roughly 3-4 years of age (Evers et al. 2010). There have been, however, reports of
juveniles in basic plumage on the breeding grounds during summer months (Ewert 1982).
1.3. FEEDING
Loons are primarily piscivorous birds that use visual, subsurface pursuit to
capture prey (Barr 1973). Preferred prey fish are those which swim somewhat erratically
such as yellow perch (Perca flavescens), pumpkinseeds (Lepomis gibbosus) and bluegill
(Lepomis macrochirus). Most prey items are consumed underwater; however larger
items are frequently brought to the surface for manipulation prior to swallowing (Barr
1973). Pursuit of fish usually occurs in shallow waters less than 5 meters deep in areas
within 50 – 150 m of shoreline, where preferred prey items are usually located (Ruggles
1994). Loons supplement their diets with aquatic arthropods and macroinvertebrates,
though underwater consumption of prey makes it difficult to quantify the proportion of
the diet that these items comprise (Gingras and Paszkowski 2006). Chicks are fed a diet
of fish and aquatic invertebrates (especially crustaceans) above the surface of the water
(Barr 1973, Alvo et al. 1988).
1.4. PHYSICAL CHARACTERISTICS
Some physical characteristics of loons are distinctive and set them apart from
other water birds. Loons are sexually monochromatic during both the breeding and non-
breeding season. They exhibit a degree of size dimorphism, with males generally
weighing 27% more than females, though this is often difficult to quantify in field
observations. The body mass of loons in Maine is between 2780-5400 grams. The legs
of loons are positioned towards the back of their bodies which aids in swimming and
underwater maneuvering, yet this positioning makes it impossible for loons to walk
4
upright on land. Instead, loons must push themselves on their ventral surface when it is
necessary to travel on land, such as during copulation, nest building, or incubation. Their
heavy body weight and shorter wing span relative to body size makes it difficult for loons
to take flight from the surface of the water. Often, they need upwards of 200 m to build
up the speed to take flight, though strong winds will greatly reduce the distance needed
for take-off. Adults can increase or decrease the amount of air in their air sacs and thus
vary the depth at which they float at the surface of the water (Evers et al. 2010).
1.5. BEHAVIOR
Loons are conspicuous birds that are known for their unique behaviors. They
spend the majority of their time floating on the surface of the water and when searching
for food they will often “peer” underwater before committing to diving after prey (Evers
et al. 2010). This peering behavior is also seen when loons are interacting with territorial
intruders. Loons will dip their heads below the surface of the water as two or more
individuals swim in a circle (pers. obs.). Aggressive interactions escalate from here and
include splash-diving (a dive beneath the surface that is more forceful than typical
foraging dives), bill striking, “penguin dancing” (the loon pulls almost the entirety of its
body out of the water and propels itself with its feet across the surface of the water),
“wing rowing” (where loons propel themselves quickly across the surface of the water
using their wings as oars, often in pursuit of a rival), and grabbing of a rival loon’s head
or bill. Once two rivals have locked bills, individuals will hit each other with their wings
and/or one rival’s head will be held underwater. Fights between conspecifics can result
in death, often by drowning or bill puncture wounds incurred when loons fight
underwater (Piper et al. 2008b, Evers et al. 2010).
5
There are a variety of behaviors performed by loons on a regular basis that are
largely involved in maintenance. These behaviors include head scratching, bathing,
stretching, preening and sleeping. Loons will perform a foot waggle, which is when one
foot is pulled out of the water and shaken and often remains out of the water, either
extended or subsequently tucked beneath the wing. This action is thought to primarily
service as a comfort movement, but it has also been speculated to serve an indirect
function in thermoregulation (Paruk 2009).
Loons are known for their loud, eerie vocalizations. Only males perform the
yodel call, which is a high amplitude call that carries a long distance (Olson and Marshall
1952) and provides information regarding a male’s motivation, condition, and identity
(Mager et al. 2007). Both males and females will vocalize using tremolos and wails, and
also create quieter sounds called “mews” and “hoots” (Mager and Walcott 2007).
1.6. BREEDING INFORMATION
Loons are monogamous on the breeding grounds and genetic studies have not
found evidence of extra-pair mating (Piper et al. 1997a). The courtship ritual of loons
consists of bill dipping, circular swimming, and synchronous diving. Common loons
have much stronger breeding site fidelity than mate fidelity (Evers et al. 2010). The pair
bonds last, on average, 5 years. If one of the pair members does not return to the
breeding site, it will be replaced through the passive occupation of another individual.
Loons will also attempt to usurp an individual from its territory though aggressive
encounters (41% of territory turnovers), which sometimes result in the death of one of the
combatants (Piper et al. 2000, 2008b). Loons are highly territorial during the breeding
season (Evers et al. 2010).
6
The female lays 1-2 eggs and both parents alternate incubation duties for 26-31
days and mutually care for the semi-precocial young (Evers et al. 2010) . The sex ratio of
chicks is very close to 50:50 (Evers 2001a, Piper et al. 2008a). Loons prefer to nest on
islands over mainland sites and will nest on artificial islands, natural islands, or on
floating bog mats. When nesting on the shoreline, loons prefer soft-edged locations on
the lee side of the lake with enough water for an underwater approach (Evers et al. 2010).
Generally, females return to nesting territories from their wintering grounds
before males (Evers et al. 2010). Territories are of three specific types: multi-lake, in
which pairs must visit more than one lake in order to meet their nutritional requirements;
single lake, in which pairs are the only loons present for a given body of water; and
shared lake, where more than one pair nests on a given lake (Piper et al. 1997a).
1.7. SITE FIDELITY
There are many advantages to returning to a breeding site year after year,
including knowledge of potential nesting sites, information regarding prey availability
and distribution, and familiarity with the presence of predators (Greenwood and Harvey
1982). Furthermore, returning to a previously used site increases chances of re-pairing
with the same partner. This may increase efficiency in nesting activities through reduced
time needed to become familiar with a mate (Rowley 1983), which may lead to increased
nesting success. Another advantage is familiarity with nearby territory holders which
could reduce the need for boundary disputes and aggressive interactions (Stamps 1987).
There have been many studies citing a positive correlation between reproductive
success and continued site fidelity (reviewed by Switzer 1997). On the contrary, poor
nesting success may cause birds to change territories and/or mates more readily the
7
subsequent year (Payne and Payne 1993). In birds, those that breed successfully are more
likely to return to a given site and have shorter dispersal distance if they do move
(Drilling and Thompson 1988). It is thought that breeding site fidelity is an adaptive
response as successful breeders can return to the area where they previously had success
and unsuccessful breeders can try elsewhere (Sedgwick 2004). It has been demonstrated
in passerines that they will use information from the previous breeding season to make
choices regarding breeding site use in the subsequent year (Doligez et al. 1999).
Both site fidelity and mate fidelity have been studied in loons. There is sex-
biased familiarity with breeding sites in loons, as only males have been found to choose
nest site locations. Therefore, it appears no benefit is gained from an increased tenure of
female loons on a territory in regards to nest site selection, and novel males experience a
“Familiarity deficit”, a reduced knowledge of successful nesting locations compared to a
long-term resident, regardless of the territorial tenure of his mate (Piper et al. 2008a).
Piper et al. (2008b) speculated that territories containing many high-quality nesting sites
(i.e. island habitats a fair distance from shoreline) would have a small familiarity deficit,
and thus there should be an increase in attempts at territorial evictions on these territories.
On the other hand, territories that lack abundant suitable nesting sites should experience
fewer eviction attempts. Alternatively, the reconnaissance hypothesis states that
territorial take-overs are more likely to occur on territories that have successfully
produced chicks the previous year, indicating that loons are monitoring the outcomes of
nesting attempts by other pairs (Piper et al. 2000).
8
1.8. MANAGEMENT AND CONSERVATION
The population growth of common loons (Gavia immer) has slowed during the
last 20 years (Grear et al. 2009). There are various explanations as to why this is
occurring, including lead poisoning from fishing weights (Pokras et al. 1998), increased
human activities on lakes (Hemberger et al. 1983), changes in land use (Lindsay et al.
2002), and increased mercury in the environment (Kenow et al. 2007). It is also possible
that the recent decrease in growth may be a result of density dependence, which is
supported in some regions by increases in aggressive interactions between nonbreeding
adults and territorial pairs (Grear et al. 2009, Piper et al. 2000, 2006).
BACKGROUND INFORMATION
1.9. RANGE EDGES
Determining the reasons for a species’ distribution can vary from quite clear, as
when a species is strongly tied to a particular resource, to incredibly complex due to the
interplay of various competing ecological and evolutionary dynamics. Studying species
on the edge of their range and understanding their distribution necessitates the
consideration of adaptive evolution and demographic processes (Holt and Keitt, 2005).
Additionally, species distributions must be considered within context of interactions with
other species in regards to predation and competition (Gaston, 2003).
Gaston (2003) outlined a three-question framework for examining species and
their associated range edges. First, one must consider the biotic and abiotic factors that
limit further spread. Examples of these factors are physical barriers or a lack of available
resources. Secondly, one must address how these factors manifest themselves in terms of
9
population dynamics (i.e. immigration, emigration, etc.). Lastly, one must consider the
genetic basis for the limit of a species’ geographic range. Using these questions as an
outline allows for a comprehensive approach to address spatial distribution.
Often, current models for determining distribution are linked solely to correlations
with ecological parameters and thus are limited in their abilities to predict ranges
(Thuiller 2003). The development of fully dynamic models linking species’ ranges to
ecological and demographic processes that occur throughout the biogeographic range are
necessary to predict the effects of habitat degradation and loss in times of a changing
climate and rapid landscape evolution (Holt and Keitt, 2005).
1.10. CLIMATE CHANGE
Climate change in the coming century is predicted to cause a suite of effects on
the global environment (IPCC, 2007). Among the projected changes are an increase in
mean surface air temperature, an increase in the concentrations of carbon dioxide in the
atmosphere, increases in concentrations of other greenhouse gases (i.e. methane, nitrous
oxide, tropospheric ozone), warming of mean sea-level temperature, and a rise in global
mean sea-level (Watkinson et al. 2004). Additionally, predicted changes in seasonality of
precipitation could greatly impact the structure of existing ecosystems (Watkinson et al.
2004).
Regional changes in New England are predicted to increase levels of
precipitation, extend drought periods, and increase air temperatures, and these changes
will likely alter freshwater ecosystems (Poff 2002). For example, a prolonged drought
can alter the movement of solutes and water to a lake, in turn modifying the lake
chemistry (Magnuson et al. 1997). All three climate divisions of Maine (Coastal,
10
Northern, and Southern Interior) are warmer and wetter than they were 30 years ago
(Jacobson et al. 2009). Increases in both air and water temperatures will most likely
affect the production, abundance, and distributions of plants in lakes and streams
(Jacobson et al. 2009). On an extremely localized level, stream flow has been linked to
regional and global climate factors (Kingston et al. 2007).
1.11. MODELING SPECIES DISTRIBUTION IN RESPONSE TO CLIMATE
CHANGE
One of the most common ways to determine possible changes in bird species’
distribution due to changing climate is through the use of bioclimatic envelope models.
These models link the current geographical distribution of a species to climatic variables
and then the future predicted location of this ‘envelope’ is used to predict where species
will occur under climate change scenarios (Heikkinen, 2006).
There are various opinions regarding the validity of modeling changes in
distribution using ‘envelope’ models, some of which support this method and others that
refute it (Pearson and Dawson 2003, Hampe 2004, Thomas et al. 2004, Hijmans and
Graham 2006, Heikkinen et al. 2006, Beale et al. 2008, Vallecillo et al. 2009). Most
researchers agree that this method is only a first step in determining the predicted changes
in distribution, and they should only be used as guidelines for further investigation. In
fact, a recent study by Beale et al. (2008) found that when they made real and artificial
‘envelope’ models for 100 different birds species, associations between species and
climate determined by climate envelope models are no better than chance (those made
artificially) for 68 of the species tested. Possible reasons for this result is the possibility
that distributions may not match climate (Beale et al. 2008) and may, in fact, be
11
explained better by land cover, biotic interactions, dispersal, or adaptive evolution (Davis
et al. 1998a, Thomas et al. 2001, Heikkinen et al. 2006). Additionally, future conditions
may include greater habitat fragmentation, species-specific responses to increased CO2
levels, soil and fire changes, and interactions of genetic differences as species are brought
together or separated (Heikkinen, 2006). These possible changes cannot be accounted for
by current envelope modeling techniques.
A better approach to modeling potential changes in species distribution is through
the inclusion of additional factors thought to influence a species’ current distribution. It
is also important to take into account the influence of species-specific interactions and
changes in human activity. Metrics that are specific to the species and can be shown to
influence the demographic processes underlying range edges would be the most valuable
for determining how changes in these parameters may alter future distributions.
1.12. LAKE METRICS PREVIOUSLY SHOWN TO AFFECT LOON
PRESENCE/ABSENCE AND PRODUCTIVITY
We included variables in our analysis (Chapter 2) that were important for loons in
smaller scale studies or variables that we hypothesized might be so on a continental scale.
As a means of testing potential drivers of range change, we included all of the following
characteristics in our nation-wide presence/absence database (and as many as possible in
the smaller datasets):
1.12.1. Lake Depth
Alvo et al. (1988) found that lake depth provided significant discrimination
between lakes where chicks were successfully raised versus where they failed. In a study
of the potential effects of climate change on freshwater systems in New England, Moore
12
et al. (1 8) determined that under a 3-5 C climate change, annual stream flow will
decrease by 21- 31%. These analyses were conducted for a doubling of atmospheric CO2
which is projected to occur in the early half of the next century (IPCC, 2007).
Additionally, predicted decreases in snowpack and increases in spring precipitation
events may lead to an earlier, yet smaller, pulse of water in spring (IPCC, 2007).
Decreases in seasonal runoff would most likely produce lower lake levels. Reservoirs
may be able to compensate for this change through alterations in dam operation; however
power companies may be forced to draw off water more rapidly during the loon breeding
season to match increased demands.
1.12.2. Lake Surface Area
Alvo et al. (1988) found that lake surface area provided significant discrimination
between lakes where chicks were successfully raised versus where they failed. In this
study, lake area was related weakly but significantly to depth, and neither depth nor area
was related to water chemistry variables. The volume of water in a lake is related to the
surface area, and runoff decreases for the reasons mentioned above, so too may lake
surface area. The trophic system of a lake is dictated by lake volume, water residence
time, and the available nutrients that flow into the lake from the surrounding drainage
basin (Wetzel, 2005). Lake trophic states may then be altered or disrupted by climate
change (IPCC, 2007), through anthropogenic changes in nutrient enrichment, or other
changes to the drainage basin (Wolfe et al. 2001). These changes have the potential to
affect the food webs upon which loons rely (Evers et al. 2010; McIntyre, 1994)
13
1.12.3. Water Clarity/Turbidity
Water clarity affects loon’s ability to see prey items such as fish and other aquatic
vertebrates (McIntyre and Barr, 1997). When water visibility is less than 1 meter,
crayfish (decapod crustacean) constitute up to a third of a loon’s diet (Barr, 1973). A
Wisconsin-based study by Lathrop et al. (1999) found that one third of year-to-year water
clarity differences in the summer were due to runoff variability. Increased nutrient input
leads to increased phytoplankton levels which, in turn, decreases water clarity. Predicted
increases in extreme rainfall events with climate change will increase the influx of
nutrients and sediments into streams and lakes, and this effect would be exacerbated if
these rainfall events occurred in areas with bare agricultural fields in the drainage basin
(IPCC, 2007). Vertical mixing explains a second third of water clarity variation.
Warmer temperatures can lead to less mixing which often results in greater water clarity
(IPCC, 2007). While turbidity will potentially increase or decrease on the local level due
to climate change, it is clear that any decreases in water clarity will result in diminished
foraging success for the common loon (Gostomski and Evers 1998).
1.12.4. Surface temperature
Blair (1992) studied the association of loon presence with 37 different lake
characteristics in New Hampshire. Surface temperature was determined to be one of the
top three predictors of presence (along with surface area and lake depth) during the
breeding season. This correlation may exist due to the effects of temperature on fish
distributions and growth and activity rates (Moyle and Cech, 2004).
Global temperatures are predicted to increase between 1- C in the next 100
years, with the magnitude of regional temperature difference predicted to be larger at
14
higher latitudes (IPCC, 2007). Increases in local temperature would logically increase
lake surface temperatures (Ficke et al. 2007). Biochemical reactions are dictated by fish
body temperature and therefore projected changes in water temperature will influence
physiological processes in fish (i.e. growth rates, metabolism, reproductive success, etc.).
Alterations in water temperature regimes are predicted to alter species abundances,
change available fish biomass, and shift current distributions (Ficke et al. 2007).
Additionally, rising water temperature will increase lake stratification duration and
strength as a warmer epilimnion increases density gradients in a way that is more
resistant to mixing (Peeters et al. 2007). Changes in stratification will decrease
availability of dissolved oxygen to the hypolymnion and alter algal assemblages, food
web dynamics, and the available suitable habitat for fish (Ficke et al. 2007). While loons
are adapted to surviving on various different prey items (Evers et al. 2010), large-scale
shifts in prey composition could potentially push loons out of once suitable breeding
lakes.
1.12.5. pH
Alvo et al. (1988) reported an inverse relationship between loon breeding success
and lake pH. McNicol et al. (1995) found that loons were less likely to nest on acidic
lakes (with pH less than 5.5), and those that did had lower reproductive success. Further,
high alkalinity lakes in Ontario were associated with successful breeding in loons (Alvo
and Berrill, 1986). Alvo and Berrill (1986) speculated that this link between productivity
and pH was due to low food levels available for the offspring (adults could forage
elsewhere when necessary). When adjusted for age, Merrill et al. (2005) found there was
a decline in food intake for loons on lakes with low pH. In this intensive study, however,
15
the reason for this decline was found to be due to reduced numbers of loon dives in acidic
lakes and not because of a reduction in provisioning time.
Contrary to these results, Badzinski and Timmermans (2006) found that there was
no correlation between loon productivity and lake acidity. They speculated that this
conflicting result was due to the predominance of lakes with neutral pH in this study
location in Nova Scotia, Canada (<10% had a pH ≤ 5.5). Parker (1 88) found that there
was no relationship between low pH lakes and loon reproductive success, however the
chicks on low pH lakes were fed smaller prey items, and the adults spent up to four times
more time feeding their offspring than adults on lakes with higher pH.
Changes in pH, acidification, and recovery of acidified lakes are projected to be
impacted by climate change through increases in atmospheric acidic deposition
(Magnuson et al. 1997, IPCC 2007). Depending on the position in the drainage basin,
lakes that receive less groundwater would be particularly vulnerable to acidification in
drier years (Webster et al. 1996). The inclusion of a larger set of lakes for analysis at a
continental scale will provide us with a more complete understanding of the effect of
acidification on loon presence and reproductive success.
1.12.6. Conductivity
Conductivity is a numerical value that corresponds to a water body’s capacity to
carry an electrical current (Wetzel, 2005). This value is directly related to the number of
dissolved ions (charged particles) in a given water body. Increases in lake pollutants,
such as sediments, inorganic elements (e.g. sodium, chlorine, sulfur), and organic
elements (i.e. decaying biological material, human made compounds) can increase
measures of conductivity, therefore acting as an indirect measure of pollution (Gleick et
16
al. 2002). Baseline measures of conductivity have the potential to indicate loon presence
on a lake. Alvo et al. (1988) found a high correlation between conductivity, alkalinity,
and pH, and these three factors together accounted for 46% of the habitat variability in
their study of loon breeding lakes in Canada (using principal component analysis). Blair
(1992), however, found no association between conductivity and loon presence/absence
on a given lake.
Human activities, such as urban or agricultural runoff, road salting, and leaking
septic waste systems are the leading contributors to increasing lake conductivity.
Additionally, new development or increased agriculture in a drainage basin can expose
new bedrock or soils and alter runoff patterns (Dale, 1997), which can increase
conductivity (Wetzel, 2000). Most alterations to landscapes occur on a regional scale and
thus the likelihood of modifications under climate change conditions is best considered
on a case-by-case basis using spatially explicit models. The proximity of a lake to large
urban centers, the feasibility of land for conversion to agriculture, or the availability of
resources (such as timber for harvesting) are all factors that determine potential land-use
changes and the ionic concentrations of runoff (Dale, 1997).
1.12.7. Dissolved Organic Carbon (DOC)
Badzinski and Timmermans (2006) found that there was no effect of DOC on
loons rearing one chick, however, adults rearing two chicks were more successful on low
DOC lakes. High concentrations of DOC (particularly on lakes with low pH) reduce
mercury availability to fish (Driscoll et al. 1995). The resulting reduced mercury risk to
loons may explain the higher success rate on lakes with higher DOC (Badzinski and
17
Timmermans, 2006). High DOC also decreases water clarity (Wissel et al. 2003),
making visual hunting more difficult for loons.
DOC is the product of the degradation of living material, and as such, it plays an
important role in the cycling of carbon in aquatic environments (Hader et al. 1998).
Projected decreases in stream flow will decrease the influx of organic material from the
lake catchment (Schindler et al. 1997). Additionally, climate change will allow for an
increased penetration of ultraviolet light (UV-B) and photosynthetically active radiation
(PAR) into lakes (Schindler et al. 1997), resulting in decreased bioavailable DOC (Hader
et al. 1998). While decreased DOC would increase loon visibility for foraging, DOC also
provides the base of the bacterioplankton food web , and changes in the phytoplankton
community would have consequences throughout the food chain (Ficke et al. 2007).
1.13. USING TRIGLYCERIDES AND FREE GLYCEROL TO DETERMINE
ENERGETIC COST
Plasma metabolites will give a snap-shot indication of energetic condition through
comparison of blood triglyceride and glycerol measures amongst individuals. High levels
of triglycerides will indicate that a loon is acquiring fat reserves whereas an elevated
level of glycerol is evidence of the use of stored fat reserves; however caution must
always be exercised when interpreting metabolite results. A loon in a more compromised
energetic state (as evidenced by higher levels of glycerol) will presumably be more
physiologically taxed and less likely to maintain healthy chicks. In Chapter 3, I analyze
loon metabolites to determine whether physiological indicators of lipid
anabolism/catabolism vary across habitat and territory types in adult loons.
18
Table 1. Descriptive Statistics for Metabolite Values. Descriptive statistics for the average
triglyceride and glycerol levels of common loons (Gavia immer) with chicks in areas of ME, NH,
MA, MT, WA during the summer of 2010 and 2011.
n mean SD
Free Glycerol
(mmol/L)
105 0.157 0.080
Triglycerides
(mmol/L)
105 0.875 0.376
Previous studies have used measures of glycerol and triglycerides to examine
energetic costs to individuals, and these findings can place our results (Table 1) in a
proper physiological context. In the remainder of this section, I attempt to demonstrate
that: 1) our findings for loons show similar triglyceride and free glycerol values to other studies,
2) triglyceride and free glycerol correlate with condition, 3) this correlation varies depending on
metabolic activity and/or food intake, 4) metabolites track condition better than mass or activity
measurements, and 5) you can use a single measure of triglycerides and free glycerol to track
condition.
As outlined by Jenni-eiermann and Jenni (1998), the physiological state in
animals is mostly a function of both feeding and metabolic rates, and the precise location
of an individual along both of these axes can be assessed by their blood metabolites and
the gross condition of their tissues which cover a continuous range: 1) feeding birds with
a mid-level metabolic rate (the resorptive state), 2) fasting birds that are inactive, 3)
fasting birds with a high metabolic rate (e.g. during migration), 4) starved birds that are
inactive, and 5) starved birds with a high metabolic rate (e.g. flight). More generally,
individuals can be assigned to one of three basic energetic states: net positive energy
budget, net negative energy budget, and stasis.
19
In our study, each loon could only be captured one time and it was important that
we could use a single measure to interpret results. Jenni-Eiermann and Jenni (1994)
designed a study to help understand the plasma metabolite readings of birds only caught
one time. In their experiment, they found that 61% of the changes in body mass were
explained by triglycerides and β-hydroxy-butyrate. Free glycerol levels were found to be
lowest when an individual’s body mass was increasing and highest when body mass was
decreasing in a study of garden warblers (Sylvia borin), however glycerol levels were of
an intermediate level when fat reserves were stable (Fig. 1; Jenni-Eiermann and Jenni,
1994). In this same study, triglyceride levels followed the opposite pattern where the
highest values occurred during fattening and the lowest values occurred during fat loss
(Fig. 1). In our study, the values we found for free glycerol are similar to the low points
graphed in the free glycerol row (Fig. 1). Lower glycerol values in the Jenni-Eiermann
and Jenni (1994) study corresponded to either stable mass or increasing mass. In their
study, increases in mass are either from an average mass to a surplus of fat, or an increase
in mass from a highly depleted state. Our values for triglycerides (Table 1) averaged
much lower than any of the values in the Jenni-Eiermann and Jenni (1994), however, we
interpret our results as being most similar to that of Group 2, letter b (Fig. 1). Among
our low levels of free glycerol, therefore, we examined differences in the values of
triglycerides as indicators of differences in energetic demand as they circulated more or
less fatty acids (bound to glycerol) in their bloodstreams.
20
Figure 1. Portion Of A Figure From Jenn-Eiermann and Jenni (1994). Relevant figure
information has been included. “Mean values (+SD) of metabolite concentrations, and
proportion of two electrophoresis peaks for the four body-mass conditions of each experimental
group. Probability values concern differences among body-mass conditions (two-way ANOVA
without replication). There were no significant added variance components among birds, except
for triglycerides of group 2 (P = 0.004) . Sample size 10 for each mean value.”
As common loons are breed on lakes and winter in the ocean, here I present a
study of other seabirds which have comparable physiological requirements to loons. In
1997, Newman et al analyzed blood from 13 species of pelagic marine birds in order to
establish baseline plasma biochemical ranges in wild Pacific seabirds. Free glycerol
levels were not reported in this study. Birds were sacrificed from the Shumigan Islands,
AK, between June 9th
and June 17th
, 1990. Species collected included ancient murrelet
(Synthliboramphus antiquus), black-legged kittiwake (Rissa tridactyl), Cassin's auklet
(Ptychoramph marmoratus), common murre (Uria aalge), crested auklet (Aethia
21
cristatell), glaucous-winged gull (Larus glaucescen), horned puffin (Fratercula
corniculata), marbled murrelet (Brachramphu marmoratus), northern fulmar (Fulmarus
glacialis), parakeet auklet (Cyclorrhynchusa psittacula), pelagic cormorant
(Phalacrocorax pelagicus), pigeon guillemot (Cepphus columba), tufted puffin (Lunda
cirrhata). Across all seabirds combined, the researchers found significantly higher
triglyceride levels in females versus males, but no significant relationships between
triglyceride levels in measures of breeding condition. The average triglyceride value for
males was 2.484 (±2.698) mmol/L and for females it was 4.177 (±5.430) mmol/L. As
there is a large amount of variance around these results, we find our triglyceride levels
fall within this range.
The correlation of triglycerides and free glycerol depend on metabolic activity
which can aid in understanding differences in physical exertion in loons. Jenni-Eiermann
and Jenni (1992) examined triglycerides and glycerol levels in the night- migrating birds
the European robin (Erithacus rubecula), garden warbler (Sylvia borin), and pied
flycatcher (Ficedula hypoleuca). In all three of these species, triglyceride and glycerol
levels were significantly higher in birds that were flying overnight than in those that
fasted overnight, because they were mobilizing fats to burn while flying (Fig. 2). The
higher plasma triglyceride and glycerol values during migration are due to the release and
transport of fatty acids from adipose tissue to flight muscles at an accelerated rate. One
hour after migratory flights, however, triglycerides remained high (as fat deposits are
being rebuilt) while glycerol is similar to fasting levels (as the rate of tissue turnover is
slow again). Our glycerol and triglyceride levels are much lower than those of the
migrating birds, although they are similar to those of the birds with lower metabolic rates
22
(the overnight fasted inactive birds or the 60 minutes after flight birds: Fig. 2). Given this
relationship, we are able to predict metabolite concentrations for loons in different
energetic conditions (assuming a similar metabolic rate).
Figure 2. Figure From Jenni-Eiermann And Jenni (1992). “Means + SDs of plasma fat
metabolite concentrations and of the percentage of fraction 1 in lipoprotein electrophoreses of
three species of birds in three physiological situations. Numbers below the bars denote the
sample sizes. Significance of the difference between adjacent samples (Mann-Whitney U-test) is
shown in asterisks: *, P< 0.05; **, P< 0.01.”
As all of the loon captures in this study were performed at night when loons are
not feeding, it is helpful to understand the dynamics between fasting and metabolite
levels. Alonso-Alvarez and Ferrer (2001) examined glycerol and triglycerides in yellow-
legged gulls (Larus cachinnans) at various stages of fasting. In this study, they found
that plasma triglycerides steadily decreased throughout fasting (over a two week period).
On the first day of fasting, gull triglyceride levels were 0.854 mmol/L (± 0.117 SD),
which is exactly our mean nightly average for loons, and these values fell to 0.605
mmol/L (0.113) after two weeks of fasting. For gulls with a restricted diet (limited
sardines fed), the triglyceride levels were similar on the first day (0.832 ± 0.056 mmol/L),
and fell even further to 0.364 mmol/L (0.050) after two weeks of dietary restriction. The
23
lowest triglyceride value in my data set was 0.120 mmol/L and so lower triglyceride
levels are thus likely predicted by decreases in energetic condition. Totzke et al. (2009)
examined the influence of fasting on herring gulls (Larus argentatus). Glycerol was not
measured in this study. Birds were starved for six days, but the initial days of the
experiment are most comparable to our values. During day 0-1 of fasting, triglyceride
levels were 1.2 mmol/L (.1) and in fasting gulls, they were also 1.2 mmol/L (.1). Similar
results were found in a second experiment. These values are higher than our mean
triglyceride values. However, this could be due to differences in extra mass in these gulls
that was not present in our sample of loons.
In species with biparental care, triglyceride values have indicated that adults can
maintain or recover (in the case of females) body mass throughout incubation. Alonso-
Alvarez et al (2002) studied changes in body mass and plasma biochemistry during
incubation in yellow-legged gulls (Larus cachinnans). Triglyceride values averaged 1.25
(+ 0.13) mmol/L for the first 10 days of incubation and 1.16 (±0.14) mmol/L during the
last ten days of incubation. Female and male triglyceride values were not significantly
different from each other and did not change significantly throughout incubation, which
indicated stable lipid reserves in the sampled birds (since body fat is correlated with
triglyceride values). Glycerol levels were not reported in this study.
At breeding season-level metabolic rates, metabolites have been shown to track
fat stores better than body mass and activity levels. Jenni and Schwilch (2001) studied
changes in metabolite levels as they varied with changes in mass in reed warblers
(Acrocephalus scirpaceus). Triglyceride levels were positively related to change in body
mass and time of day, while β-hydroxy-butyrate plasma levels were negatively related to
24
these same parameters (Fig. 3). Therefore, levels of plasma β-hydroxy-butyrate and
triglycerides can be used to estimate body mass change in birds only captured one time
(Glycerol levels were not reported in this study). Even though both of these metabolites
reflect change in body mass, triglycerides demonstrate changes in fat stores more closely,
as they are directly related to fat deposition, unlike β-hydroxy-butyrate, which is related
more to transitions from one state to the other. Thus, triglyceride levels are better
predictors of changes in lipids stores than overall change in body mass. Body mass and
activity level were not correlated with triglyceride levels, and so within the range of body
masses in the study, stores of fat did not influence triglyceride levels, which indicates that
heavier birds may use alternate energy forms for overnight fasting and rely less on the
catabolism of lipids (thus their lower β-hydroxy-butyrate levels by morning).
Figure 3. Figure From Jenni and Schwilch (2001). “Relationship between hourly change in
body mass of reed warblers since early morning (ΔMASS-B) and triglycerides (a) or β-hydroxy-
butyrate levels (b). The metabolite values displayed have been standardized to 6 h after lights on
according to the models.”
At low metabolic rates (like those experienced during the extended inactivity
associated with incubation), triglyceride levels correlate positively with energetic
condition. In 2010, Bauch et al. examined triglyceride, uric acid, and cholesterol levels
in incubating common terns (Sterna hirundo). In this study, triglyceride levels were
25
measured during three different periods of the incubation (early, middle, and late) and
two levels of breeding experience (experienced and inexperienced). Glycerol values
were not examined in this study. In 2006, the highest triglyceride values were measured
during the early incubation period and ranged from 1.00 mmol/ L (inexperienced
breeders) to 1.26 mmol/L (experienced breeders). Regardless of experience level, birds
in higher energetic condition (prior to incubation) exhibited higher triglyceride levels,
and individuals in higher condition (experienced breeders) also exhibited higher levels
than more inexperienced individuals. In 2007, there was more variation between
triglyceride levels, but again the greatest variation was in the earliest part of incubation.
The values in this year ranged from 1.10 mmol/L (inexperienced breeders) to 1.69
mmol/L (experienced breeders).
1.14. SYNOPSIS
In the two chapters that follow, I investigate the relationships between
environmental characteristics and common loon occupancy, productivity, and
physiological condition along the southern edge of the breeding range. I aimed to
determine factors that influence distribution and understand the links between climate
change and range edges. My overarching goal was to determine if climate change will
impact loon demography and the location of the southern range edge through habitat
preferences, changes in energetic state, and reproductive success on previously viable
lakes.
The scope of my analysis took place on varying geographic scales, with
narrowing focus. The largest occupation model in Chapter 2 included lakes across the
United States portion of the historical breeding range (south to the maximum extent of
26
the Laurentide Ice Sheet). Reproductive success was analyzed within the Northeastern
US across the region currently occupied by loons. Behavioral analysis encompassed a
portion of lakes in Maine, New York, Vermont, and Massachusetts, where more detailed
observations have been conducted. Finally, energetic measures were studied on a subset
of loons sampled from lakes in New Hampshire, Western Maine, Massachusetts,
Washington and Montana.
Our results from the large-scale occupancy model identified lake salinity, acidity,
and sulfate levels as significant predictors. Similar factors drove loon presence/absence
in New England (lake salinity and alkalinity) as well as a unique parameter, lake surface
area. Loon productivity, on the other hand, was best predicted as either high or low using
the size of both the lake and drainage basin. Lake surface area and is thus good
predictors of both loon distribution and loon productivity, and are thus also likely to be
useful in predicting range shifts in the future. These outcomes suggest that future range
alteration for loons due to climate change is likely to be more sensitive to annual adult
survival (which will influence breeding ground settlement patterns) than environmental
factors encountered on the breeding grounds.
For our physiological examination of range contraction we used measures of
triglycerides (as free glycerol was very low and showed little variation, as we would
expect for fasting birds maintaining mass with relatively low to mid metabolic rates).
Blood triglyceride concentrations suggest that the number of chicks, breeding latitude,
and daily temperature are important predictors of loon energy expenditure. Specifically
we found that: 1) loons with two chicks expend more energy than those with one, 2)
loons near the southern range edge expend more energy to produce a given brood size
27
than those nearer the range center and, 3) birds breeding in warmer temperatures expend
more energy than those in cooler temperatures (controlling for year, territory type, and
calendar date, free glycerol levels, size-corrected body mass, and longitude). Species-
specific physiological thresholds of temperature, are known to limit the distribution of a
wide variety of organisms (Nickerson et al. 1988, Walther et al. 2002, Pörtner 2002), and
it is reasonable therefore to assume that we can detect initial differences in triglyceride or
glycerol values under thermal conditions that are less extreme. This initial variance could
then potentially act as a harbinger of larger scale population changes.
Together, the findings from both of my chapters suggest that there are scalar and
biological differences in detectability of factors that will cause range contraction under
climate conditions. Clearly, occupancy, productivity, and physiological state were best
predicted by different environmental characteristics. Of the predictors related to loon
distribution, only SO4 was predicted to be influenced by climate change. Our large-scale
analysis found that human impacts will have the greatest initial impact on distribution
and that of the predictors related to productivity, none were predicted to change with
climate change. My model did not include all possible factors that could influence loon
presence/ absence or productivity. For example, I did not test underlying bedrock or
other large geographic features of the landscape which may have also helped explained
patterns in limnology and loon distribution. I also did not test the social and behavioral
factors that might limit the dispersal of loons. The larger scale models are helpful for
identifying possible trends in occupation across wide spatial scales, which is the scale of
climate change that may cause shifts in loon distributions.
28
With my physiology model, I detected influences on energetic demand (which
could manifest themselves as loss in productivity) and found these factors to be tied to
temperature and to warming trends. The physiology model was able to detect factors
related to climate change because of the influence of temperature on metabolic rate. The
physiological model is perhaps most informative model in terms of understanding of
climate change as it captures the physical, behavioral, and environmental forces acting on
a given individual.
29
CHAPTER 2
MACHINE LEARNING TECHNIQUES: A TOOL FOR UNDERSTANDING
COMMON LOON (GAVIA IMMER) BIOGEOGRAPHY AND
PRODUCTIVITY IN AN ERA OF CLIMATE CHANGE
2.1. ABSTRACT
Climate change has the potential to shift and restrict ranges for a suite of species.
The birds of the boreal ecosystem, like the common loon (Gavia immer), may be
particularly at risk given the changes predicted for this biome. Current range models for
this iconic water bird predict that large sections of the United States may lose the loon in
the next 100 years, but these models are based on habitat correlations and not the
demographic mechanisms that will actually produce the change. The primary goal of our
research was to understand the factors contributing to the vulnerability of loons to
climatic change at multiple scales. We applied a recursive partitioning technique to
analyze loon presence/absence in 288 lakes across the southern edge of their North
American distribution using 112 abiotic and landscape-level factors. The resulting binary
tree (“decision tree”) classified lakes into groups based on the probability of loon
presence, while maximizing homogeneity within the resultant two nodes. The most
significant splits in the cross-validated tree were created using lake salinity, acidity, and
sulfate levels. We employed similar methods to compare loon occupancy and seasonal
fecundity at a smaller scale (New England) to elucidate potential demographic
mechanisms of loon persistence. Results from twenty potential predictors suggest that
processes similar to the continental model are also driving loon presence/absence in New
30
England (lake salinity, alkalinity, and lake surface area). Loon productivity, on the other
hand, was best predicted as either high or low using the size of both the lake and drainage
basin. Lake surface area and characteristics of the drainage basin are thus good
predictors of both loon distribution and loon productivity, and are thus also likely to be
useful in predicting range shifts in the future. As few (if any) of the predictors of
productivity in the best decision trees are likely to change dramatically with climate,
these outcomes suggest that future range alteration for loons due to climate change is
likely to be more sensitive to annual adult survival (which will influence breeding ground
settlement patterns) than environmental factors encountered on the breeding grounds.
2.2. INTRODUCTION
2.2.1. Climate Change and Freshwater Communities
Climate change is predicted to cause numerous effects on the global environment
in the coming century (IPCC, 2007). Among the projected changes in the United States
are increases in mean surface air temperature, an increase in the concentration of carbon
dioxide in the atmosphere, increases in concentrations of other greenhouse gases (e.g.
methane, nitrous oxide, tropospheric ozone), warming of mean sea-level temperature, and
a rise in global mean sea-level (Watkinson et al. 2004). Additionally, predicted changes
in the seasonality of precipitation could greatly impact the structure of existing
ecosystems (Rodenhouse et al. 2007). In New England, precipitation is predicted to
increase, drought periods are expected to extend, and air temperature is projected to
increase, and these changes will likely alter freshwater ecosystems (Magnuson et al.
1997, Poff 2002, Jacobson et al. 2009) and local hydrologic regimes (IPCC 2007).
31
These anticipated abiotic transitions will undoubtedly affect the distribution and
characteristics of the biota, as well. Certainly, changes in air and water temperature can
alter the production, abundance, and distributions of aquatic organisms (Jacobson et al.
2009). Climate change has been shown to impact avian distributions through range shifts
(Hitch and Leberg 2007), contractions (Virkkala et al. 2008), and expansions (Martinez-
Morales et al. 2010). Accurately predicting changes in the distributions of these or other
organisms using abiotic drivers, however, has proved elusive (Davis et al. 1998a, 1998b,
Thomas and Lennon 1999, Beale et al. 2008, Williams and Jackson 2012).
Determining the controls on a species’ distribution can vary from the obvious
(e.g. resource specialists) to the complex due to the interplay of competing ecological
and evolutionary dynamics (Gaston 2003). Studying species on the edge of their range,
however, helps us understand controls on organismal distributions and necessitates the
consideration of adaptive evolution and demographic processes (Holt and Keitt 2005,
Holt et al. 2005). Many existing models predicting climate-induced changes in animal
distributions are based on bioclimatic envelopes (contemporary correlations between a
species’ distribution and other taxa or abiotic conditions) and thus are limited in their
ability to predict changes accurately (Thuiller 2003, Heikkinen et al. 2006). In a majority
of cases, distributions may be explained better by biotic interactions, limits to dispersal,
adaptive evolution, or other alterations of basic demographic parameters (Davis et al.
1998a, Thomas et al. 2001, Heikkinen et al. 2006).
One of the largest flaws of bioclimatic envelope modeling is the assumption that
current correlations with distribution will hold under future conditions. This assumption
may be safer when the environmental correlations are tied directly to demographic
32
processes (Chase et al. 2005, Seavy et al. 2008) or average individual condition, rather
than tied indirectly to the distributional patterns that result from these lower order
processes. Demographically explicit models, however, are data intensive and infeasible
for modeling most continental-scale distributions. Here we suggest a hybrid approach,
where we first develop an envelope model for the southern breeding distribution of the
lake-residing common loon, Gavia immer, in North America, and then we validate this
model by comparing the best indicators of range edge at the continental scale to those
identified by a smaller-scale, demographically explicit model (chick production across
New England). Additionally, we use hypothesis testing to determine how climate change
is most likely to alter future distributions, by A) identifying a set of environmental
variables that have the potential to change over the next century and then B) testing the
ability of these variables to explain the current loon distribution. While this will not
identify non-linear range responses to changes in these variables outside the current
observed range of variation, it will allow us to determine whether candidate predictors of
the current range edge are also candidates for change themselves. This coupled
demographic validation and hypothesis-driven approach can identify drivers of
demographic change at continental scales, even for ecological futures without a direct
contemporary analog. For the common loon in particular, this represents a discrete step
forward in identifying the likelihood of range change.
2.2.2. Machine Learning: Using Decision Trees and Random Forests to Determine
Variable Importance
We used a machine-learning approach as it presents some distinct advantages for
predicting range limit drivers over the more traditional general linear or generalized
33
linear models used in bioclimatic envelope models. Classification and regression-tree
analyses (or decision trees) are becoming a more widely used method for modeling
species distribution due to their ease of interpretation and ability to simultaneously
process various explanatory variable types (e.g. categorical, numeric, and continuous).
Decision trees are most noteworthy in their ability to investigate data with correlated and
complex relationships (De’ath and Fabricius 2000). While more traditional methods of
modeling a species’ presence versus absence (e.g. logistic regression) quickly outgrow
the ability to handle multiple predictors and their interactions, classification trees
consider each predictor as a potential variable for splitting the data (O’Connor and
Wagner 2004). Decision trees also describe a series of nested, hierarchical thresholds, an
approach that may reflect the actual physiological and behavioral processes that
determine range edges.
In decision trees, the entire dataset represents what is called the “root node”.
Each subsequent split is referred to as a “node”, and the final nodes (after which there is
no further partitioning of data) are referred to as “leaves” or “terminal nodes”. The
candidate predictor variable that explains the greatest deviance in the response variable
produces the first node. Using a recursive partitioning method, data are then split with
the remaining explanatory variables to maximize heterogeneity between the left and right
branches and homogeneity within each resultant branch (and decrease the Gini index,
which is a measure of impurity; further explained below) (Crawley 2007).
Despite their merits, decision trees have a tendency to over-fit the data and can be
unstable, whereby a small change in the data can result in a large change in the outcome
of the tree (De’ath and Fabricius 2000). This problem can be avoided by cross-
34
validating the tree using a cost-complexity measure (known as “pruning” the tree) to
decrease the misclassification rate, and both problems can be avoided by comparing
decision-tree results to those produced by random forests, a resampling technique that
addresses the instability in machine learning. Random forests can also assess the strength
of individual variables by comparing their performance across multiple trees constructed
from subsets of the data through the use of a composite Gini index, an impurity criterion
(Cutler et al. 2007). The Gini index is higher for the parent nodes than the descendent
nodes, thus high summed Gini indexes (the Gini Importance Measure) across the random
forest represent variables that occurred more commonly in higher nodes. Further,
random forests are not prone to over-fitting like individual decision trees can be.
2.2.3. General Approach
We included seven variables in our analysis of the southern edge of the loon
breeding range that were previously reported as important in smaller scale studies and
several dozen additional variables that we hypothesized might be important on a
continental scale. Previously reported predictors of loon distributions or reproduction
included lake depth and lake surface area (Alvo et al. 1988), water clarity (Barr 1973,
Gostomski and Evers 1998, Lathrop et al. 1999, Evers et al. 2010), water surface
temperature (Blair 1992), pH (Alvo et al. 1988, Parker 1988, McNicol et al. 1995, but see
Badzinski and Timmermans 2006), conductivity (Alvo et al. 1988, Blair 1992), and
dissolved organic carbon (Badzinski and Timmermans 2006). We collected values for all
of our candidate variables from sites across the current southern breeding limit in North
America (including parts of Washington, Idaho, Montana, Minnesota, Wisconsin,
Michigan, New York, Maine, Vermont, New Hampshire, and Massachusetts; Evers et al.
35
2010). If any of these metrics influence demographic processes along this range edge,
they are potentially more informative for predicting future loon distributions than values
from range-wide envelopes, as range contraction or expansion occurs presumably as a
function of conditions along edges. This more narrow spatial focus should increase the
chances that we identify patterns important for predicting changes in the southern limit of
distribution.
We used machine learning techniques to model (1) loon presence/absence on a
continental scale as a function of landscape and lake-level predictors and (2) productivity
on a regional scale to determine if similar variables influenced distribution and
productivity. We also discuss the potential of highly ranked model predictors to change
with climate over the next century. The development of models that link species’ ranges
to ecological and demographic processes is necessary to predict changes in organism
distributions under changing climates and rapid landscape evolution (Holt and Keitt
2005). Though our results are specific to loons, this paired continental and regional
comparison, underpinned by specific climate change hypotheses, is applicable to any
species on any scale and may be an improvement upon traditional envelope modeling.
2.3 METHODS
2.3.1. Methods Overview
We constructed a continental-scale decision tree of the environmental
characteristics driving the common loon southern boundary in North America. To
validate the demographic underpinning of this tree, we subsequently compared the results
of this model to a decision tree predicting the productivity of loons in a long-term
demographic study across the New England region (1987 – 2010). As the productivity
36
model was created using fewer predictors than the continental model, we validated this
model with a presence/absence model using a reduced number of predictors, as well. All
decision tree analyses were substantiated using a random forest approach..
2.3.2. Continental-scale Range Descriptors
For the continental-scale decision tree, we used data collected by the
Environmental Protection Agency for the 2007 National Lakes Assessment (NLA)
Report (http://water.epa.gov/type/lakes/lakessurvey_index.cfm). The goals of the NLA
study were to assess the biological health of lakes of the United States and to determine
baseline measures for future studies. The study included 909 lakes selected using a
probability-based design. Lakes smaller than 4 hectares were not included in the
analysis, which coincidentally corresponds with the size of the smallest lake reported to
support chick production by common loons (4.05 hectares; McIntyre 1994). We chose
lake data from this study due to its spatial scale, the consistency of collection (all data
collected within the summer of 2007), the precise data collection procedures, and the
large number of parameters measured at each lake.
We collected loon presence/absence data for 288 NLA lakes from state agencies
and non-governmental organizations that have gathered monitoring information within 5
years of 2007 (Table 2). Loons were deemed “present” or “absent” on a given lake using
a consistent hierarchical classification process (Fig. 4). Any lakes that could not be
confidently determined as having loon “presence” or “absence” under these criteria were
excluded from the analysis (final lakes included in analysis: Fig. 5).
37
Table 2. Data Sources. Organizations that contributed common loon (Gavia immer)
presence/absence data to this study (data collected within five years of 2007). State Organization(s)
Maine Maine Audubon, Biodiversity Research Institute
Massachusetts MassWildlife, Massachusetts Division of Conservation and
Recreation
Michigan Michigan Natural Features Inventory, Biotics 4- Michigan’s Natural
Heritage Database; Biodiversity Research Institute, Lake Superior
State University
Minnesota Northland College
Montana Montana Common Loon Working Group, Montana Fish, Wildlife &
Parks, ND: Northern Prairie Wildlife Research Center
New Hampshire Loon Preservation Committee; Idaho: Idaho Fish and Wildlife
Information System
New York Biodiversity Research Institute’s Adirondack Center for Loon
Conservation, Wildlife Conservation Society’s Adirondack
Program
Vermont Vermont Center for Ecostudies
Washington Northeast Washington Common Loon Observation Project Daniel
and Ginger Poleschook (Biodiversity Research Institute)
Wisconsin LoonWatch, Sigurd Olson Environmental Institute, Northland
College
Wyoming Wyoming Game and Fish Department., Loon Lake Loon
Association, Wyoming Game and Fish Department
Figure 4. Determining Loon Presence/Absence. Hierarchical classification process for
determining common loon presence, absence, or exclusion from analysis for lakes in the National
Lakes Assessment Database (2007), 2005-2009.
38
Figure 5. Lakes Used In Analysis. Location of the maximum extent of the Laurentide ice sheet
(blue line) and locations of lakes (green = present, white = absent; n = 288) included in nation-
wide analysis of variables thought to influence common loon distribution across their southern
breeding range edge, 2005-2009.
To identify the environmental characteristics that best predict the current southern
range boundary of common loons, we gathered information from the NLA dataset for
U.S. lakes within a belt that varied in width across the country, but included the areas
once covered by the Laurentide ice sheet which began retreating between 18 and 13 ka
(Fig. 6; Dyke and Prest, 1987). Lakes created during the Late Wisconsian and Holocene
times are more similar to each other than to those outside the range of this ice sheet
(Dyke and Prest 1987), and there is historical evidence that the breeding range of the
common loon included lakes south to this glacial boundary within the last millennium
(Evers et al. 2010). We did not include lakes that currently support loons within small
disjunct populations in the mountainous west south of the continuous range (e.g.
39
Wyoming), because our primary goal was to predict the continuous southern range
boundary, and we suspect different settlement processes are at play in this region (Evers,
2010). We also excluded lakes from the plains states (ND, SD, and Western MT),
because there is no evidence that these lakes were ever occupied by loons within the last
millennium (Evers et al. 2010). This exclusion only eliminated one currently occupied
breeding lake from this entire region.r determining lake inclusion
Figure 6. Map For Determining Lake Inclusion. Map demonstrating inclusion process for
lakes in nation-wide presence/absence analysis, 2005-200 . Numbers represent a “region”
variable included to account for regional differences in factors predicting presence/absence. The
“x’s” in WY and ND represent small and isolated population of loons in this region and thus
these regions were excluded from analysis due to our goal of predicting factors effecting range
edge along a continuous area. The blue line represents the maximum extent of the Laurentide ice
sheet, which began retreating between 18 and 13 ka (Dyke and Prest, 1987).
40
Table 3. Variables in National Presence/Absence Analysis. Variables included in
nation-wide analysis of common loon presence/absence along their southern range edge
in the United States, 2005-2009. (* signifies variables combined to make new category;
^ scored as: None/Low/Moderate/Heavy ; ** scored as Low=1, Medium=3, High=5)
Watershed Landuse Metrics
Variable name:
Percent of Watershed as:
PCT_OPENH20_BSN
Open water
PCT_BARREN_BSN
Barren land
PCT_GRASS_BSN
Grassland/herbaceous
PCT_SHRUBLAND_BSN
Shrub/scrub
BSN_PCT_SHRB_GRS
*Non-forest/crops
(shrub/scrub, grassland/herbaceous)
PCT_PASTURE_BSN
Pasture/hay
PCT_CROPS_BSN
Cultivated crops
BSN_PCT_FARM
*Farm (pasture/hay and cultivated crops)
PCT_DEVHIGH_BSN
High intensity development
PCT_DEVLOW_BSN
Low intensity development
PCT_DEVMED_BSN
Medium intensity development
PCT_DEVOPEN_BSN
Open space development
BSN_PCT_DEVEL
*Developed (high intensity, low intensity,
medium intensity, open space)
PCT_DECID_BSN
Deciduous forest
PCT_MIXED_BSN
Mixed forest
PCT_CONIF_BSN
Evergreen forest
BSN_PCT_FOREST
*Forested (evergreen, deciduous, mixed)
PCT_WDYWET_BSN
Woody wetland
PCT_EMHERBWET_BSN
Emergent herbaceous wetland
BSN_PCT_WETLD
*Wetland (woody wetland, emergent
herbaceous wetland)
Lake Buffer Landuse Metrics
Variable name:
Percentage of buffer as:
PCT_BARREN_BUFR_CONV
Barren land
PCT_CONIF_BUFR_CONV
Evergreen forest
PCT_DECID_BUFR_CONV
Deciduous forest
PCT_MIXED_BUFR_CONV
Mixed forest
BUFR_PCT_FOREST_CONV
*Forest (mixed, deciduous and evergreen)
PCT_DEVHIGH_BUFR_CONV
High intensity development
PCT_DEVLOW_BUFR_CONV
Low intensity development
PCT_DEVMED_BUFR_CONV
Medium intensity development
PCT_DEVOPEN_BUFR_CONV
Open space development
41
Table 3. Continued
Lake Buffer Landuse Metrics
Variable name:
BUFR_PCT_DEVEL_CONV
*Developed (high intensity, low intensity,
medium intensity, open space)
PCT_GRASS_BUFR_CONV
Grassland/herbaceous
PCT_SHRUBLAND_BUFR_CONV Shrub/scrub
BUFR_PCT_SHRB_GRS_CONV *Non-forest/crops (shrub/scrub,
grassland/herbaceous)
PCT_CROPS_BUFR_CONV
Cultivated crops
PCT_PASTURE_BUFR_CONV
Pasture/hay
BUFR_PCT_FARM_CONV
*Farm (pasture/hay, cultivated crops)
PCT_EMHERBWET_BUFR_CON
V Emergent werbaceous wetland
PCT_WDYWET_BUFR_CONV
Woody wetland
BUFR_PCT_WETLD_CONV
*Wetland (woody wetland, emergent
herbaceous wetland)
Lake Physical Characteristics
Variable name:
Variable Description:
DEPTHMAX
Maximum observed lake depth (m)
ELEV_PT
Site elevation (meters) from National
Elevation Dataset
LAKEAREA
Lake polygon area (km^2) from National
Elevation Dataset
LAKEPERIM
Lake polygon perimeter (km) from NHD
LEVEL_CHANGES
Elevation change (m) of lake level
MACRY_FLOAT
Emergent/floating coverage (% lake area)
MACRY_SUBMERGED
Submergent/floating coverage (% lake area)
PCT_EMRG_SUB_VEG
*Percent of submergent/floating,
emergent/floating coverage
SLD
Shoreline devel't index
(LAKEPERIM/(2*sqrt(LAKEAREA*pi))
LONG
Longitude
BASINAREA_HA
Watershed area
REGION
Region 1: NJ, VT, NH, NY, NJ, PA, RI, CT,
MA; Region 2: MI, OH, IL, IN, eastern WI;
Region 3: Western WI, IO, MN; Region 4:
WA, MT
42
Table 3. Continued
Lake Chemical Properties
Variable name:
Variable Description:
ANC
Gran ANC (ueq/L), Acid Neutralization
Capacity (ANC)
CL Chloride (ueq/L)
COLOR
Color (PCU)
COND
Conductivity (uS/cm @ 25 C)
CYANDENS
Cyanophyta density (#/cm2)
DO_FIELD
Field dissolved oxygen (mg/L)
DOC
Dissolved organic carbon (mg/L)
H
H+ from PH_LAB (ueq/L)
K
Potassium (ueq/L)
MCYST_TL_UGL
Total microcystin concentration (ug/L)
MG
Magnesium (ueq/L)
NA
Sodium (ueq/L)
NH4
Ammonium (ueq/L)
NO3
Nitrate (ueq/L)
NTL
Total nitrogen (ug/L)
OH
Hydroxide from PH_LAB (ueq/L)
PH_FIELD
Field pH from Profile DO data
(pH measured at first non-zero
depth unless only depth was zero)
PTL
Total phosphorus (ug/L)
SIO2
Silica (mg/L SiO2)
SO4
Sulfate (ueq/L)
TEMP_FIELD
Field water temperature (oC)
TURB
Turbidity (NTU)
Lake impacts and disturbances
Variable name:
Variable Description:
AGR_CROPLAND
^Agricultural cropland disturbances obs’d
AGR_FEEDLOT
^Agricultural feedlot
disturbances obs’d
AGR_LIVESTOCK
^Agricultural livestock disturbances obs’d
AGR_ORCHARDS
^Agricultural orchards disturbances obs’d
AGR_PASTURE
^Agricultural pasture disturbances obs’d
AGR_POULTRY
^Agricultural poultry disturbances obs’d
AGR_WITHDRAWL
Agricultural water withdrawl
AGR_IMPACT
*Summmed agricultural stressors
IND_SCORE
**Summed industrial stressor score
MAN_SCORE
**Summed lake management stressor score
43
Table 3. Continued
Lake impacts and disturbances
Variable name:
Variable Description:
REC_FILMS
^Recreational surface films, scums, slicks
disturbances obs'd
REC_MARINAS ^Recreational marinas disturbances obs’d
REC_PARKS
^Recreational parks disturbances obs’d
REC_PRIMITIVE
^Recreational primitive parks, camping
disturbances obs'd
REC_RESORTS
^Recreational resorts disturbances obs’d
REC_TRAILS
^Recreational hiking trails disturbances
observed
REC_TRASH
^Recreational trash/litter disturbances obs’d
REC_IMPACT
*Summed recreational stressors
RES_BRIDGES
^Recreational bridges/causeways
disturbances
observed
RES_CONSTRUCTION
^Recreational construction disturbances
obs’d
RES_DUMPING
^Residential dumping disturbances obs’d
RES_LAWNS
^Residential maintained lawns disturbances
obs’d
RES_PIPES
^Residential pipes, drains disturbances obs’d
RES_RESIDENCES
^Residential residences disturbances obs’d
RES_ROADS
^Residential roads disturbances obs’d
RES_SEWAGE
^Residential sewage treatment disturbances
obs’d
RES_IMPACT
*Summed residential stressors
Shoreline characteristics
Variable name:
Percentage of shoreline:
RCH_AGRICULTURE
agriculture
RCH_BARE_GROUND bare ground
RCH_FOREST
forest
RCH_GRASS
grass
RCH_SHRUB
shrub
RCH_WETLAND
wetland
RCH_DEVELOPMENT
development (Res & Urban)
RCH_SHOREMODS
modified (docks, riprap)
RCH_ALTERED
*human modification (development,
modified)
44
We compared the explanatory power of 112 lake characteristics, which we
characterized into six general categories: watershed land-use metrics, lake buffer land-use
metrics, lake physical characteristics, lake chemical properties, lake impacts and
disturbances, and shoreline characteristics (complete list: Table 7). We removed one lake
from consideration because of NLA data that were significantly outside the range of all
other lakes (Grubb’s test for one outlier, p < 10-16
). We also included two variables to
allow differential range determination by region (longitude and a categorical variable for
the four glacial lobes identified in Figure 4). To validate this approach for identifying
regional differences, we additionally constructed a decision tree for New England using
the same predictors from our continental model. The results from this tree agreed with
those from the continental model; therefore, we do not include further explanation of that
model in this chapter.
When creating the continental classification tree, we set the minimum number of
observations in a node to 30 before each split could be attempted and required the
resulting split to reduce the overall lack of fit by a cost complexity factor of 0.001. The
full tree explaining loon presence and absence was cross-validated ten times (using the
“xval” command within the “rpart” statement of R), and we then used the complexity
parameter to prune each tree to a size that minimized cross-validation error (Breiman et
al. 1984). We validated the pruned tree by running a random forest of 1000 trees from
the original dataset to assess the stability of our results. The “out-of-bag” (OOB) error
rate did not indicate that the random forest would have benefited from the addition of
more trees. Because there is no currently accepted standard for the number of predictors
that should be attempted at each node, we varied our values from 1 to 112 (in increments
45
of 10) and used the number of predictors that yielded the lowest OOB error rate (“mtry”
= 50; OOB error rate = 9.98%). Lastly, we created variable importance plots for all
predictor variables using both the decrease in mean Gini index value and the mean
decrease in model accuracy after the removal of each variable (calculated as the
normalized difference between the classification accuracy of the number of correctly
classified samples before the removal of a given variable from the number of correctly
classified samples after permutation, divided by the number of OOB samples for that
tree). This value was calculated for each of the 1000 trees in the random forest and its
average thus corresponds to the importance of a given variable in a model. We then
conducted partial dependence plots to explore the importance of top-ranked variables
within the random forest models.
2.3.3. Regional-scale Range Descriptors
The majority of lakes for which we had detailed demographic information were
not contained within the NLA database. We therefore constructed a decision tree using
the lakes of VT, NH, NY, and ME, but only considered the 20 variables that were
available for our lakes with demographic information as well (Table 4). These lake data
were compiled from the Environmental Protection Agency Wildlife Database Search
Engine (see “WDSE” for further information below).
46
Table 4. Variables in New England Presence/Absence Analysis. Variables included in
the analysis of common loon presence/absence along their southern breeding-range edge
in New England states, United States, 2005-2009.
Variable name Variable Description
BASINAREA_HA Watershed area
ELEV_PT Site elevation (meters) from National Elevation Dataset
DEPTHMAX Maximum observed lake depth (m)
LAKEAREA Lake polygon area (km^2) from National Elevation Dataset
LAKEPERIM Lake polygon perimeter (km) from NHD
CHLA Chloride (ueq/L)
PH_FIELD Total phosphorus (ug/L)
ANC Gran ANC (ueq/L), Acid Neutralization Capacity (ANC)
TURB Turbidity (NTU)
CL Chloride (ueq/L)
NO3 Nitrate (ueq/L)
SO4 Sulfate (ueq/L)
CA Calcium (ueq/L)
MG Magnesium (ueq/L)
NA Sodium (ueq/L)
K Potassium (ueq/L)
COLOR Color (PCU)
COND Conductivity (uS/cm @ 25 C)
DO_FIELD Field dissolved oxygen (mg/L)
PTL Total phosphorus (ug/L)
2.3.4. Regional Productivity Descriptors
We then constructed a decision tree to describe loon productivity across the
Northeastern United States. The Biodiversity Research Institute (BRI) has been
collecting data on loons throughout this region since 1987. The most continuous and
rigorous data are from the mountains of western Maine, beginning with work on 4 lakes
(12 loon territories) in 1993 (Dodge Lake, Quimby Lake, Rangeley Lake, and Round
Lake). Today, data are collected on 179 territories over 38 lakes in western Maine.
Additional data were included from New Hampshire (Loon Preservation Society),
Vermont (Vermont Center for Ecostudies), Massachusetts (MassWildlife, Massachusetts
47
Division of Conservation and Recreation), and New York (Biodiversity Research
Institute’s Adirondack Center for Loon Conservation and The Wildlife Conservation
Society’s Adirondack Program). Information from these sources was used to construct
our New England productivity decision tree.
We modeled the number of chicks surviving per pair (after four weeks of age)
using environmental lake characteristics found in the Environmental Protection Agency
Wildlife Database Search Engine (WDSE: http://oaspub.epa.gov/aed/wildlife.search). We
included 20 candidate variables from the WDSE (Table 5) that were measured
consistently across all lakes with productivity information. Loon productivity and lake
characteristics (when there were multiple entries for either) were averaged over each
decade for each lake (e.g. 1 80’s, 1 0’s, 2000’s). Averaging data by decade allowed us
to examine lakes on a fine spatial scale while correcting for individual loon-intrinsic or
seasonal variance in productivity or lake characteristics. The productivity means were
calculated across all territories within a lake, resulting in a continuous response variable
of average number of chicks per pair per lake for the decade, thus averaging out the
differences in productivity caused by either different individuals on different territories or
different territory holders across years. Tree construction operated similarly to the
previous models. We ran cross-validation ten times and plotted the cross-validation
error. We plotted the mean squared error (MSE) and total decrease in node impurity
(from splitting on a given variable) and our regression trees were substantiated with this
random forest model. We validated both of these models with random forests in the same
way we did our continental scale model.
48
Table 5. Variables in New England Productivity Analysis. Variables included in
analysis of common loon productivity along their southern breeding-range edge in New
England states, United States, 1980-2010.
Variable Name Variable measurement
Loon Decade Decade loon captured
watershed area acres Watershed area, acres
Lake surface area acres Lake surface area, acres
max lake depth m Maximum lake depth, meters
elevation m Elevation, meters
shoreline length m Shoreline length, meters
ANC
Acid neutralizing capacity, mg/L
calcium
Calcium, mg/L
chlorophyla Chlorophyl A, ug/L
chloride
Chloride, mg/L
color
Color, SPU
DOX
Dissolved Oxygen, mg/L
MAGNESIUM Magnesium, mg/L
NITRATE
Nitrate, mg/L
PH
pH, pH units
PHOSPHORUS Phosphorus, mg/L
POTASSIUM Potassium, mg/L
SECCHI
Secchi, meters
SODIUM
Sodium, mg/L
SCONDUCT Specific conductance, uS/cm
SULFATE
Sulfate, mg/L
TEMP
Temperature, degrees Celcius
TURBIDITY Turbidity, NTU
We created all of the classification and regression trees using R version 2.14.1,
Package rpart version 3.1-50 (Therneau et al. 2012). We created the random forest using
R version 2.14.1, Package randomForest version 4.6-6 (Liaw and Wiener 2012).
2.4. RESULTS
2.4.1. Continental-scale Range Descriptors
The pruned, continental-scale decision tree used five of the 112 potential variables
across seven terminal nodes and had a residual mean deviance of 0.12 (Fig. 7). The
49
retained variables included lake water Na+ ion levels (ueq/L), lake surface area, SO4
(ueq/L) concentration, and lake elevation above sea level. For lakes with Na+ ion levels
greater than or equal to 144.3 ueq/L, loon presence was predicted positively by lower SO4
levels (threshold >= 67.68 ueq/L) and negatively by higher SO4 concentrations in the
lake. For lakes with Na+ of < 144.3 ueq/L, loons were absent on small lakes (threshold =
17.43 hectares), but in larger lakes loons were present on lakes at elevations greater than
529.2 meters above sea level and absent at lower elevation lakes.
Figure 7. United States Classification Tree. Pruned classification tree for common loon
presence/absence along the southern range edge, 2005-2009. Predicted values for splitting
decision (i.e. thresholds) and number of cases are presented at each node (absences / presences).
Predicted values of occupancy (0= absent, 1 = present) and number of cases are presented at each
terminal node. Each node (i.e. split) is shown as an inequality. If the inequality is true, follow
the left branch; if not proceed to the right branch.
Mean decreases in model accuracy and Gini indices for variables from each of the
random forest permutations (Fig. 8) largely agreed with our full, pruned tree, ranking Na+
and SO4 concentrations as the first and third most important variables, respectively. Lake
surface area also ranked high in importance for both the full tree and random forest
50
(ranked fourth and fifth by the Gini indices and mean decreases in accuracy respectively).
Chloride ion concentration was the second most important predictor in the random forest,
although it did not appear in our final classification tree. Scree plots of our two
importance rankings suggest either five (Gini Indicies) or six (Model Accuracy)
important variables overall, and include regional differences (longitude) and the percent
of mixed deciduous and coniferous forest in the drainage basin as important predictors
along with lake chemistry and size. Lake elevation was the tenth highest ranked, but was
not as strongly supported by the random forest models as the other top-ranking variables
identified by the continental classification tree (and likely correlates with forest cover in
the drainage basin).
The partial dependence plots of the six highest ranked predictorseach showed a
variety of threshold shapes for loon presence averaged over the effects of all other
variables (Fig. 9). At lower values of sulfate (SO4), chloride (Cl-), and sodium (Na
+), the
logit of probability of presence was high and then it decreased sharply (Fig. 11). An
increase in longitude (moving east) corresponded with an increase in the logit probability
of loon presence until - 4 longitude, after which the logit probability decreased steeply
until - 2 longitude. The logit probability of loon presence increased rapidly across small
lakes, and then leveled out after lakes larger than 100 hectares. The probability of loon
presence was low for values of mixed forest cover in the watershed until about 9%, after
which it increased rapidly.
51
Figure 8. Random Forest Analysis For United States. Mean decrease in accuracy and mean
decrease in Gini Index of first 33 variables for random forest model of presence/absence across
the continental United States using 112 predictive variables, 2005-2009. Variables higher on the
list are more important to explaining the response variable than those lower on the list. To
determine the mean decrease in accuracy, the prediction error on OOB data is subtracted from the
OOB error after permuting each predictor variable. These differences are averaged over the trees
(normalized by standard deviation of differences). The decrease in node impurity is the total
decrease in impurity from splitting on a given variable. To determine the mean decrease in Gini
Index (another impurity criterion), a score is assigned to a variable each time is used as one of the
primary splits in a tree. Gini Indexes are higher for the parent nodes than the descendent nodes,
thus high summed Gini indexes (the Gini Importance Measure) across the random forest
represent variables that occurred more commonly in higher nodes.
52
Figure 9. Dependence Plots for United States Analysis. Partial dependence plots give a
graphical representation of the marginal effect of a variable on the probability of presence or
absence, 2005-2009. These are partial dependence plots for the top six most important variables
for a continental analysis of common loon distribution along their southern range edge in the
United States.
2.4.2. Regional-scale Range Descriptors
Our final classification tree predicting loon presence or absence in New England
using the 20 WDSE candidate predictors consisted of 2 terminal nodes with a residual
mean deviance of 0.06 (Fig. 8). Again, Cl- concentration, an index of pH (acid
neutralization capacity instead of sulfate in this model), and lake surface area were
important variables for predicting loon distribution. Chloride concentration split the data
in half at a value of 158.3 ueq/L. At levels higher than this, the acid neutralization
53
capacity of the lake positively predicted presence (threshold = 754 ueq/L). At lower Cl-
concentrations lake surface area positively predicted loon presence (threshold = 33.9 ha).
Figure 10. Classification Tree For New England. Full and “pruned” (pruning is not justified
by cost-complexity analysis) classification tree: common loon presence/absence model for New
England states using a reduced number of predictive variables (20), United States, 2005-2009.
Predictors used in this model are the same as those used in the analysis of loon productivity in
New England. Predicted values for splitting decision (i.e. thresholds) and number of cases are
presented at each node. Predicted values for occupancy (0= absent, 1 = present) and number of
cases are presented at each terminal node. Each node (i.e. split) is shown as an inequality. If the
inequality is true, follow the left branch; if not proceed to the right branch.
The plotted mean decreases in accuracy and Gini indices for the variables in the
model showed that elevation, Na+ levels, and Cl
- levels were the most important variables
(the first splits of the trees, Fig. 11). We saw similar patterns for Na+ and Cl
-
concentrations in this smaller scale, reduced dataset as the continental scale tree where
each had low and abrupt thresholds for loon absence (negative relationship). Lake
magnesium, elevation, maximum depth, and surface area all had sharply increasing
relationships with the logit probability of presence (Fig. 12).
Tree
|CL>=158.3
ANC< 754 LAKEAREA< 0.3386
039/39
0
32/5
030/0
12/5
1
7/34
07/3
10/31
54
Figure 11. Random Forest For New England Analysis. Mean decrease in accuracy and mean
decrease in Gini Index of all 20 variables for random forest model of presence/absence in New
England, United States using 20 predictive variables, 2005-2009. Variables higher on the list are
more important to explaining the response variable than those lower on the list. For a full
description of the indices, see Fig. 8.
COLOR
NO3
PTL
CHLA
DO_FIELD
TURB
BASINAREA_HA
SO4
K
PH_FIELD
CA
COND
LAKEPERIM
ANC
DEPTHMAX
LAKEAREA
MG
CL
NA.
ELEV_PT
0.0 0.5 1.0 1.5 2.0
MeanDecreaseAccuracy
NO3
PTL
COLOR
DO_FIELD
CHLA
TURB
BASINAREA_HA
K
SO4
CA
COND
PH_FIELD
ANC
MG
DEPTHMAX
LAKEAREA
LAKEPERIM
CL
NA.
ELEV_PT
0 1 2 3 4 5
MeanDecreaseGini
pa.rf
55
Figure 12. Dependence Plots For New England Analysis. Partial dependence plots give a
graphical representation of the marginal effect of a variable on the probability of presence or
absence, 2005-2009. These are partial dependence plots for the top six most important variables
for the New England analysis of common loon distribution along their southern range edge in the
United States.
2.4.3. Regional Productivity Descriptors
Our full regression tree for the analysis of productivity in the New England
dataset used eight of the 20 candidate variables (eleven nodes) to construct the initial
classification tree; however, the pruned tree only retained two of these variables:
watershed size and lake surface area (Fig. 13). Lakes with smaller drainage basins
(threshold = 639.0 hectares) had lower loon productivity than lakes in larger drainage
0 1000 2000 3000
-0.2
0.2
NA.
0 500 1500 2500
-0.4
0.0
0.4
ELEV_PT
0 1000 2000 3000
-0.4
0.0
0.4
CL
0 1000 3000 5000-0
.20.1
MG
0 10 20 30 40 50
-0.2
0.1
LAKEAREA
0 10 20 30 40 50
-0.3
0.0
DEPTHMAX
56
basins, and these lakes had the lowest predicted productivity of the three terminal nodes
(1.2 chicks per pair per lake). Large lakes (threshold = 54.2 hectares) in larger drainage
basins had the highest productivity (1.8 chicks per pair per lake). tree for productivity model
Figure 13. Regression Tree For Productivity Model. Pruned regression tree: common loon
productivity model for loons in New England, United States, 1980-2010. Predicted values for
splitting decision (i.e. thresholds) and number of cases are presented at each node. Predicted
values for productivity (number of chicks per loon per decade) and number of cases are presented
at each terminal node. Each node (i.e. split) is shown as an inequality. If the inequality is true,
follow the left branch; if not proceed to the right branch.
For our random forest validation, our OOB error rate for our model was 13.2%,
which explained 42.4% of the variance. Drainage basin area, lake surface area, and
shoreline length were the highest ranked variables to split the data (Fig. 14). We were
unable to create partial dependence plots for the model as our original data contained too
many missing values (the tree itself is robust to these missing values (Breiman 2001) but
the partial dependence plotting procedure is not).
Pruned Regression Tree
|watershed.area.acres>=1579
Lake.surface.area.acres< 134
1.299n=126
1.18
n=94
1.65
n=32
1.347n=12
1.831n=20
57
TURBIDITYLoon.DecadechloridePOTASSIUMANCcalciumNITRATESULFATEMAGNESIUMDOXelevation.mSODIUMSCONDUCTSECCHIPHPHOSPHORUSchlorophylamax.lake.depth.mTEMPcolorshoreline.length.mLake.surface.area.acreswatershed.area.acres
0 5 10 15 20
%IncMSE
TURBIDITYLoon.DecadecalciumANCPOTASSIUMSODIUMchlorideSULFATENITRATESECCHIMAGNESIUMchlorophylaPHDOXSCONDUCTTEMPPHOSPHORUScolorelevation.mmax.lake.depth.mshoreline.length.mLake.surface.area.acreswatershed.area.acres
0 1 2 3 4 5
IncNodePurity
pa.narf
TURBIDITYLoon.DecadechloridePOTASSIUMANCcalciumNITRATESULFATEMAGNESIUMDOXelevation.mSODIUMSCONDUCTSECCHIPHPHOSPHORUSchlorophylamax.lake.depth.mTEMPcolorshoreline.length.mLake.surface.area.acreswatershed.area.acres
0 5 10 15 20
%IncMSE
TURBIDITYLoon.DecadecalciumANCPOTASSIUMSODIUMchlorideSULFATENITRATESECCHIMAGNESIUMchlorophylaPHDOXSCONDUCTTEMPPHOSPHORUScolorelevation.mmax.lake.depth.mshoreline.length.mLake.surface.area.acreswatershed.area.acres
0 1 2 3 4 5
IncNodePurity
pa.narf
Figure 14. Random Forest For Productivity Analysis. Percent increase in mean squared error
and increase in node purity all 20 variables for random forest model of productivity (continuous
response variable) in New England, United States using 20 predictive variables, 1980-2010.
Variables higher on the list are more important to explaining the response variable than those
lower on the list. These measures are analogous to the mean decrease in accuracy and Gini
Indices used for a binomial response variable.
58
2.5. DISCUSSION
2.5.1. Continental-scale Range Descriptors
Lake Na+ concentration, Cl
- , and SO4 were the most important predictors of loon
presence across the loon southern range edge in North America. The initial node with
Na+
identified 67% percent of the lakes without loons (tree: Fig. 9), and eliminated lakes
both within and outside the modern loon range. Our random forest analysis were
convergent with our pruned classification tree findings regarding the importance of Na+
(ranked most important variable in random forest; Fig. 8). Kuhn et al (2011) modeled
presence/absence in the New Hampshire and found that loons appeared to avoid human-
altered landscapes with high road densities and high levels of agricultural activity. A
similar process may be at work on a continental scale, as our results could be explained
by either a direct choice in Na+ levels, or the absence of other large landscape-level
disturbances, such as agriculture, development, or livestock grazing, which may alter lake
chemistry and increase Na+ levels (Dale 1997).
Lake Na+ concentration was the best indicator of continental loon distribution in
our random forest model (Fig. 10) and in our classification tree model. We do not
suspect this finding is due to an intolerance to Na+, as loons winter in marine
environments and have adaptations (e.g. large nasal glands, specialization of the kidneys)
to eliminate excess salt; however, further investigation into this realm is warranted. We
hypothesize instead that loons may be sensitive to the trophic effects of increased salinity
in freshwater ecosystems. Aside from natural variation in bedrock and net evaporation,
differences in lake Na+ levels are generally attributed to the runoff of road salt and
chemical deicers, which can elevate both Na+ and Cl
- levels in lakes (Rosenberry et al.
59
1999, Ramakrishna and Viraraghavan 2005) and suggests that loons may be less likely to
occur on lakes with high amounts of deicer application in the drainage basin. Across the
nation about 11 million tons of chemical deicers and road salts are spread each year
(Keating 2004), 85% of which are used in the mid-Atlantic, Great Lakes, and New
England states (Kaushal et al. 2005). This practice may be limiting the current
distribution of loons in these high-use regions. With climate change, interactions
between weather and human responses may alter this limitation. The average area of
snow cover in North America has decreased by 0. million square miles since the 1 0’s,
with the lowest value in 2006 (Armstrong and Brodzik 2006, Robinson and Heim Jr.
2006) and these trends are predicted to continue (Barnett et al. 2005). If the southern
distribution of loons is indeed limited by the ecosystem effects of road deicing, a warmer
world with less snow and ice may actually allow loons to recolonize some or all of their
former range. Alternatively, deicers may be a general index of human land use, as Cl-
concentration in lakes has also been linked to human disturbance in the drainage basin
(Herlihy et al. 1998). Under this scenario, we would not predict loon range expansion in
the near future, as human development is unlikely to reverse in the mid-Atlantic and
Midwest states.
Recent changes in the southern limit of the loon range support our initial
hypothesis. Historically, loons have been reported to breed in locations further south
than their current distribution in areas of northeast California, northern Iowa, southern
Minnesota, northern Illinois, southern Wisconsin, northern Indiana, southern Michigan,
northern Ohio, and northeast Pennsylvania where there are no modern records of
breeding (Evers et al. 2010). Loons began breeding in northern Massachusetts
60
approximately 40 years ago and in parts of Pennsylvania 30 years ago, both of which
represent either range expansions or a recolonization of their former range in these areas
(Evers et al. 2010) and may suggest a role for deicer application in current loon range
limitation.
Our variable dependence plots from the random forest model suggested a more
abrupt threshold for lake chemistry variables (such as Na+) than other landscape level
factors (Fig. 9). This pattern, however, may be attributed to correlations with other ions
that we did not test. For example the restriction of lake outflow creates a more saline
environment and increased levels of a large suite of ions (Wetzel 2005). Among the
twenty-four ions we tested (Table 8), however, Na+ repeatedly outperformed the others.
It is important to note that lakes with the highest Na+ concentrations were largely found
across the southern breeding range, suggesting that an untested correlate of latitude may
also be responsible for this pattern.
Another significant, positive predictor of loon occupancy was lake surface area.
Previous studies have analyzed lake surface area effects on loon occupancy and loon
reproductive success. Alvo et al (1988) found that lake surface area provided significant
discrimination between lakes where chicks were successfully raised versus where they
failed. In this study, lake surface area was related weakly but significantly to depth, and
neither depth nor surface area was related to water chemistry variables (as found in other
studies; Blenckner 2005). If lake surface area is driving the patterns with occupancy
directly, it is unlikely to change dramatically with climate change, although changes in
precipitation may change lake areas locally.
61
Loons may be sensitive to landscape-level forest changes, as three forest cover
variables were determined important in our top eight variables in our random forest
analysis (including one variable that our scree plot suggested was of the most important
variables). Climate change is expected to impact forest ecosystems in inconsistent ways
on varying spatial scales, with changes in disturbance regimes (Dale 1997) and plant
species ranges (Matthews 2004). A more immediate and likely change in many areas
across the northern United States over the next few decades will be from increased
population growth, local urbanization, and changes in ranching and farming practices
(Dale 1997). These land-use changes threaten to alter existing forested landscapes, which
may impact lake suitability for loon persistence. The percentage of mixed forest cover in
the drainage basin, however, primarily delineated northern lakes from southern lakes, and
therefore it is possible that the relationship we detected is driven by an untested factor
that also correlates with latitude (e.g. air temperature, day length, ice coverage).
While we would have ideally validated the results from our nationwide model
with a nationwide demographic model, this was not feasible due to incomplete data
across the entire southern range, a common constraint of demographically driven range
models. The strength of our approach, however, is that it allows for demographic
analysis in a portion of the range where these data are available. Creating an occupancy
model of the entire southern range allows for a comparison with the predictors from this
model with the variables that predict productivity in a portion of their southern range.
Further, creating an occupancy model for this same reduced area assures that any
differences detected are not due to a difference in scale rather than a difference in
mechanism.
62
2.5.2. Regional-scale Range Descriptors
To test whether the same factors influencing loon distribution on a continental
scale were operating on a regional scale, we analyzed data only for the lakes in New
York and New England states using fewer predictors. The variables retained (Cl-, acid
neutralizing capacity, and lake surface area) in the regional model were similar to those
retained on the national scale.
Cl- is a major anion in saline lakes (Wetzel 2005), and as the compound Na
+Cl
- is
a common road deicer, the identification of Cl- in this model may be analogous to the
importance of Na+ in our continental analysis. While coastal lakes can receive marine-
derived Cl- ions from atmospheric transport (Wetzel 2005), Cl
- increases are generally
due to pollution from road salt and municipal wastewaters (Sonzogni et al. 1983, Kaushal
et al. 2005).
A second variable, acid neutralizing capacity (ANC), was retained in our model
(ANC was a candidate predictor in the NLA dataset: Table 7). Total alkalinity levels are
a chief determinant of ANC (Wetzel 2005), and high alkalinity lakes are often more
biologically productive and less susceptible to the effects of acid precipitation (Driscoll et
al. 2003). Loons were largely present, therefore, on more productive lakes with higher
buffering capacities. The strength of the association between loon presence and ANC
value was not strong enough to be selected as a top predictor in the nationwide model, yet
in New England where lakes are categorized as “acid-sensitive” by the EPA (Driscoll et
al. 2003), the strength of this association was much more prominent. As our small-scale
models did not include lakes to the south of the current loon range boundary, however,
63
the inclusion of ANC in this tree may indicate that there are landscape characteristics that
predict unoccupied lakes within the range that are not predictive of the range edge itself.
While ANC appears to predict local lake occupancy and not larger scale range
edges, its local impact may increase with a changing climate. Although declines in
atmospheric sulfur dioxide emissions over the last several decades have helped to
increase ANC levels in lakes (Driscoll et al. 2003), climate-related events are reversing
this process (Laudon et al. 2004) via increased atmospheric acidic deposition (Magnuson
et al. 1997, IPCC 2007). Depending on position in the drainage basin, lakes that receive
less groundwater will be particularly vulnerable to acidification in drier years (Magnuson
et al. 1997). Predicted increases in summer drought intensity and frequency are expected
to increase acidic episodes in water bodies and cause declines in ANC (Laudon et al.
2004). Alterations in ANC, alkalinity, and lake pH could all have detrimental effects on
loon productivity and persistence on specific lakes within the loon range (McNicol et al.
1995).
2.5.3. Regional-scale Productivity Descriptors
Decadal productivity averages in New England were positively associated with
both drainage basin area and lake surface area. While lake surface area is unlikely to
change dramatically with climate change, lake surface area can indicate risks to nesting
success in the form of flash floods (which could inundate eggs) or droughts (which will
make it difficult for loons to push themselves across land to get to their nest) because of
correlations between lake size and position in the watershed. These potential climate
induced disturbances to nest success are, in fact, encountered to a lesser degree today
during extreme rainfall events or on reservoirs that draw water off quickly (Desorbo et al.
64
2007). Loon populations have persisted in most areas that experience these disturbances,
despite the increased challenge of successfully producing chicks (Evers et al. 2010).
Drainage basin size may be important due to its similar relationship to risk of lake level
variation or to other correlates with the land- cover of the drainage basin (e.g. forest
cover). Because drainage basin size and lake size were the most important factors in
predicting reproductive success, we are encouraged by the fact that these are not factors
that are likely to alter with climate change scenarios. If, however, these two variables are
important because of links with water-level variability, changes in precipitation regimes
may influence loon productivity by increasing or decreasing such risk.
While other studies have found patterns between loon productivity and the
variables we rejected (e.g. water clarity, water surface temperature, and dissolved organic
carbon), most of these studies were on a smaller scale. Undoubtedly, these correlations
between loon productivity and lake characteristics do exist, and may be important on a
smaller scale or within a single drainage basin. On a regional scale, however, the
individual effects did not predict the difference between occupied and unoccupied lakes.
2.5.4. Validating Envelope Models with Hierarchical Decision Trees
Our goals were to identify variables that were important in determining
distribution and productivity and distinguish which variables were important in one
realm, the other, or both. Our smaller scale presence/absence model identified subsets of
the continental drivers of loon range edge and thus relationships appear to vary across
space, but continental scale decision trees capture smaller-scale relationships. Lake
salinity (measured by Na+ or Cl
-), acidity (measured by SO4 or ANC), and surface area,
were all important in determining loon presence at both a continental and regional scale.
65
Our productivity model validated the demographic underpinning of two variables as
loon range predictors: lake surface area and drainage basin size. Forest composition of
the drainage basin was a consistent predictor of loon presence at both regional scale and,
to a lesser degree, the continental scale. Forest composition is driven by variables that
are scale dependent (e.g. substrate, climate, graving, site history, etc. vary on a local
versus site scale (Corney et al. 2006). Although we did not directly test forest
composition in our productivity model, the only drainage basin variable we did test,
drainage basin size, was an important predictor of loon productivity. This underscores
the demographic importance of the drainage basin to loons and suggests a possible
demographic mechanism for its power as a range predictor.
Our method of pairing our occupancy model with a demographic model made us
more confident in our occupancy model results. The productivity model rejected lake
salinity and acidity as predictors. While these two predictors appeared important at both
the regional and continental scales, we found no ties with offspring production. This
does not mean, of course, that there is no demographic underpinning to these two
variables. They may influence settling rate (immigration) or apparent site fidelity
(emmigration and survival). On the other hand, they may be spurious due to correlations
with another demographically important variable. Our approach allowed us to reject their
linkage with productivity, and further investigations could test for linkages with other
demographic mechanisms. Without the hierarchical validation, however, we would not
have been able to distinguish between the two predictors that were related to productivity
and the two predictors that were not.
66
2.5.5. Implications for Climate Change on the Range of the Common Loon
Our overall findings regarding the probability of loon persistence within the
United States are promising. All of our demographically validated variables that explain
loon distribution are unlikely to change with climate, although we need a better
understanding of how drainage basin composition and lake size influences productivity to
be confident. Of the range predictors that were not related to loon productivity, two were
unlikely to change greatly due to climate (salinity and acidity/alkalinity). In both of these
cases, however, we need to be conservative in our predictions until we understand why
these characteristics predict the current loon range. If they are not validated through a
demographic mechanism other than productivity, these relationships may be spurious and
their predictive ability for future range shifts severely limited. Smaller scale studies have
suggested demographic linkages between loons and acidity/alkalinity, but we recommend
additional large-scale investigations to validate this lake characteristic at a continental
scale. Overall, however, we believe our generalized approach of a continental envelope
model followed by smaller scale machine learning techniques to validate the
demographic underpinnings of important variables represents a distinct step forward in
developing range models that will accurately predict range shifts with landscape and
climate change. In this case study we were able 1) to reject or support the productivity
underpinnings of range predictors from an envelope model and 2) reject the ability of a
number of variables, which were previously identified as important for loon demography,
to predict the current range of the common loon. This approach should work equally well
with many species to identify variables that are both accurate predictors of current ranges
and tied to demographic mechanisms likely to lead to range changes in the future.
67
It is important to recognize that our analysis only addresses loon persistence in the
United States during a portion of their annual cycle. Loons that breed in the continental
United States spend the entirety of their winter on the Atlantic or Pacific Oceans, or, less
commonly, on unfrozen freshwater systems near the ocean (Evers et al. 2010). Future
changes in ocean conditions that lower winter survival or energy stores required for
migration and subsequent breeding may have larger impacts on the common loon
breeding range than changes directly to breeding lakes themselves.
68
CHAPTER 3
BLOOD METABOLITES AS INDICATORS OF BREEDING HABITAT
QUALITY: A SENSITIVE METHOD TO PREDICT RANGE
ALTERATION UNDER CLIMATE CHANGE?
3.1 ABSTRACT
Climate change has the potential to shift and/or restrict species’ ranges. While
previous studies have generally failed to predict range alteration successfully, individual
energetic condition may offer more potential because it can identify covariance between
energetic condition and breeding habitat quality. We used blood metabolites and
behavioral observations to evaluate the energetic costs of common loons (Gavia immer)
breeding on lakes across a gradient of environmental, spatial, and behavioral conditions.
Using these physiological and behavioral endpoints we created a framework for
evaluating the effect of environmental characteristics on the value of breeding locations
along the species’ southern range edge (ME, NH, MA, MT, WA). Using samples
collected over two years (n=97), blood triglyceride concentrations suggest that the
number of chicks, breeding latitude, and daily temperature are important predictors of
loon energy expenditure. Specifically we found that: 1) loons with two chicks expend
more energy than those with one, 2) loons near the southern range edge expend more
energy to produce a given brood size than those nearer the range center and, 3) birds
breeding in warmer temperatures expend more energy than those in cooler temperatures
(controlling for year, territory type, and calendar date, free glycerol levels, size-corrected
body mass, and longitude). We suggest that as environmental conditions change in the
69
coming years, blood metabolites offer a promising predictor of population collapse along
range boundaries.
3.2 INTRODUCTION
Climate change is predicted to cause a range of effects on animal populations
(IPCC 2007). In birds, the potential for range shifts, expansions, and contractions has
already been realized by numerous species (Watkinson et al. 2004, Waite and Strickland
2006, Virkkala et al. 2008, Barbet-Massin et al. 2009, Martinez-Morales et al. 2010).
Some of the hypothesized mechanisms behind these shifts include changes in food
availability, phenological alignment, and thermoregulatory challenges (Brown et al.
1999, Walther et al. 2002).
One of the most common ways to predict how climate will alter bird species’
distributions is through the use of bioclimatic envelope models. These models link the
current geographical distribution of a species to climatic variables, and the predicted
locations of these variables are then used to forecast the species’ occurrence (Heikkinen
et al. 2006). Most researchers agree that this method is only a first step in determining
the predicted changes in distribution, and it should only be used as a guideline for further
investigation (Pearson and Dawson 2003, Hampe 2004, Stanley and Royle 2005, Hijmans
and Graham 2006, Heikkinen et al. 2006, Vallecillo et al. 2009) as distributions may not
be limited simply by climate (Beale et al. 2008). Models that link species’ ranges to
explicit ecological and demographic processes have proven more useful in predicting the
effects of habitat degradation (from climate or other rapid landscape changes) on animal
distributions (Holt and Keitt 2005, Byrd 2012, Chpt. 2). Metrics that influence the
70
demographic processes underlying range edges are the most valuable for determining
how alterations in these parameters may change future distributions.
In order to link an organism’s distribution with an individually based mechanism
of distribution change, it is important to examine the relationship between physiology and
environment. Linking physiology to microclimatic data in a scalable way is the key to
understanding a species’ distribution (Hodkinson 1999). Species-specific physiological
thresholds of temperature, for example, are known to limit the distribution of a wide
variety of organisms (Nickerson et al. 1988, Walther et al. 2002, Pörtner 2002), and it is
reasonable therefore to assume that we can detect initial differences in metabolic
expenditure under thermal conditions that are less extreme than those needed to cause
local extirpation. This initial variance could then potentially act as a harbinger of larger
scale population changes.
Body mass is a widely used estimate of energy stores as it is noninvasive and easy
to measure, and changes in body mass can give an indication of changes in energy stores
(Jenni-Eiermann and Jenni 1994). This method of determining stores of fat reserves,
however, has shortcomings. These include the need to recapture animals (e.g. energy
stores might influence recapture ability) and the potentially adverse affects of capture on
fat acquisition (Jenni-Eiermann and Jenni 1994).
Blood metabolites are one potentially useful link between environmental
conditions and individual energetic expenditure that can shed light on an individual’s
ability to deal with changing environmental conditions (Brown et al. 1999). Plasma
metabolites are the products of intermediary metabolism and are measured in the
bloodstream between their origin within an organ and their delivery to another location in
71
the body (Jenni-Eiermann and Jenni 1998). Therefore, concentrations of plasma
metabolites relate to both the current energetic demand of an individual and the
individual’s ability to meet that demand. Very high triglyceride levels, for instance, can
indicate that an individual is increasing caloric intake or fat reserves (which may suggest
that it is more buffered from environmental changes; Quillfeldt et al. 2004). Conversely,
triglycerides are hydrolyzed during fasting (to increase fatty acid and glycerol
availability), and thus lower triglyceride levels and elevated glycerol typify fasting (or
possibly starvation) in birds (Carey 1996). Overall, plasma metabolites can be viewed as
correlates of the net balance of stored energy from the preceding hours (Jenni-Eiermann
and Jenni 1998).
We examined free glycerol and triglyceride concentrations in the blood of
brooding common loons (Gavia immer) to assess the energetic demand of birds in
different environments along the species’ southern range extent. We predicted that we
would detect low glycerol values with little variation in our population of successfully
breeding loons (captured at night during fasting). Free glycerol has shown no nightly or
diurnal trend in birds fasting through the night (Jenni-Eiermann and Jenni 1997) and
concentrations were generally low in a study of garden warblers (Sylvia borin) when an
individual’s body mass was increasing or when fat reserves were stable (Jenni-Eiermann
and Jenni 1994). In this same study, triglyceride levels followed the opposite pattern
where the highest values occurred during fattening and the lowest values occurred during
fat loss. Because our loons were successfully breeding, we assumed that they were in
relatively high condition (low glycerol and higher triglycerides). Because we caught
them all during their nightly fast, we expected varying concentrations of triglycerides as a
72
function of net energy assimilation and use from the previous day. Therefore, we
predicted that we would see differences in triglyceride levels which would indicate
energetic state. This same link between fat reserves and triglyceride levels has been
found in two studies of fasting gulls (Totzke et al. 1999, Alonso-Alvarez and Ferrer
2001). Further, a study of reed warblers (Jenni and Schwilch 2001) concluded that
triglyceride levels were better predictors of changes in lipid stores than overall change in
body mass. Finally, while a study of thirteen species of seabirds near Shumigan Island,
Alaska found no differences in triglyceride levels in various stages of breeding condition,
all of our samples were collected from loons that were brooding chicks to control for
variation due to breeding stage.
In the past several decades, the population growth rates of common loons (Gavia
immer) have declined in several regions of the northern United States (Sidor et al. 2003,
Karasov et al. 2007, Evers et al. 2008, Grear et al. 2009) and the collapse of populations
in this southern portion of the range has been predicted by a bioclimatic envelope model
(Matthews 2004). The best predictors for both loon distribution and seasonal
productivity in North America, however, are fairly robust to near-term changes in climate
and are more sensitive to human disturbance (e.g. forest cover in the drainage basin, Na+
levels in a lake, lake surface area: Byrd 2012, Chpt. 2). We expect, however, that if
range contraction along the southern extent of the loon distribution occurs due to changes
in the breeding environment, the cost of reproduction will increase prior to the
appearance of measurable changes in demographic variables such as offspring production
or bird abundance. Demographic models agree that changes in factors affecting adult
survival are more likely to cause population change than variation in fecundity directly
73
(Grear et al. 2009). Evaluating the energetic demand of breeding adults will help to
understand the conditions within the current breeding range that are most likely to affect
population viability and cause range contractions.
To construct a more physiologically mechanistic model of common loon
persistence, we examined the way in which environmental and social parameters at
different scales varied with the blood triglyceride concentration of individuals on the
breeding grounds, controlling for glycerol concentrations. These relationships may then
identify environmental variables that impact adult survival and control both loon
population viability and drive range changes. Here, we examine common loon
triglyceride levels to link environmental variance with energetic condition during what
we believe is one of the most energetically demanding portions of the loon life cycle,
chick rearing.
3.3 MATERIALS AND METHODS
3.3.1. Study Population, Field Methods, and Lab Assays
Loon captures took place between May and August of 2010 and 2011. Blood
samples were collected in Maine (n= 77), New Hampshire (n= 22), New York (n= 4),
Washington (n = 4), and Montana (n= 4). Breeding adults with chicks were captured at
least 30 minutes after evening civil twilight using both spotlighting and playback
recordings (Evers 1992, 2001b). To identify individuals in the future, loons were fitted
with unique color bands in addition to a USGS issued metal band. All individuals were
of known sex, were caring for chicks, and were captured at night (i.e. were fasting: Evers
et al. 2010). Standard morphometric measurements were taken (e.g. tarsus width, weight,
bill length) to create an index of body size (Senar and Pascual 1997). Within fifteen
74
minutes of capture, we collected a blood sample in a heparinized capillary tube following
the venipuncture of the metatarsal vein. The samples were kept on ice until centrifuged
(within six hours) at 6000 rpm for 10 minutes. Plasma was frozen until laboratory assays
were performed.
We conducted triglyceride/free glycerol assays to determine the levels of free
glycerol, total glycerol, and triglycerides from blood samples. Assays were conducted in
accordance with the Triglyceride/Free Glycerol Assay Protocol from the Holberton
Laboratory of Avian Biology (Holberton 2008). Using a colorimetric enzymatic endpoint
assay we measured plasma glycerol levels (GPO-Trinder Assay, kit 337; B Sigma-
Aldrich, St. Louis, Missouri). Plasma samples were run in duplicate using 96-well plates
and read on a SpectraCount reader (Packard, Palo Alto, California). We used a
commercial standard of known triglyceride levels and created a standard curve to
standardize our results.
3.3.2. Behavioral Observations
We recruited biologists from four Northeastern states to gather behavioral
observations of 113 loon territories. In particular we were interested in the frequency of
escalation of territorial interactions. We expected that one of the most energetically
expensive activities in which territorial birds could engage would be territorial disputes.
During these interactions, territorial birds (both males and females) may chase intruders
around the lake at high speed while “rowing” with their wings on the surface of the
water. They may also physically grapple, beating their wings against one another and
attempting to spear their opponents with their bills (Piper et al. 2000, 2008b, Evers et al.
2010). Not all interactions escalate to this level, however, and interactions may end after
75
as little as a vocal call and some slight changes in posture. Territorial defense in birds
may account for 0.9 to 33.1% of daily energy expenditure during the height of the
breeding season (Walsberg 1983). While 41.5% of territory acquisitions in loons occur
through passive takeover (desertion or death of previous territory holder), the same
percentage of acquisitions occur through the active displacement of a member of the
resident pair or through earlier arrival time at the territory (Piper et al. 2000). Thus,
defense of a territory has the potential to require a large portion of a loon’s seasonal
energetic budget.
We hypothesized that the likelihood of an intense escalation was at least partially
a function of the energetic condition of the territory holder. Biologists across New
Hampshire, New York, Vermont, Massachusetts, and Maine recorded the frequency of
territorial intrusions and the level of escalation of an interaction by the territory holder
(“passive”: intruder is present but territory holder is not currently engaged in an
aggressive interaction or “active”: territory holder is having an aggressive interaction
with intruder including peering, circle dancing, pursuing, directly physically contacting
intruder, directly vocalizing at intruder, or other) from May-August of 2010 and 2011.
Loons were observed from the lakeshore or from a kayak and the first behavior observed
by the researcher were recorded. Behaviors were recorded separately for both members
of the pair and marked individuals were differentiated when possible.
3.3.3. Statistics
In order to determine the average triglyceride and free glycerol levels in our
samples, we used Microsoft Office Excel (2007) to find the mean and standard deviation
of our measures.
76
We identified 17 candidate variables a priori that made biological sense based on
field observations, exploratory data analysis, and published literature. Candidate
predictors of loon condition explained variation in the immediate landscape (2), in the
larger region (2), in the local social environment (3), in the timing of the sample (6), and
among individuals (4).
We identified four spatial variables. Our two landscape-scale effects included
drainage basin area and lake surface area, both of which are landscape-level predictors of
loon productivity (Byrd 2012, Chpt. 2). Lake surface area may correlate with several
social parameters, as well. Nocera and Burgess (2002) found that lake surface area was
positively correlated with dive-pause times and variation in dive-pause times, which they
speculated was because loons on larger lakes may be more apt to encounter intruders
(through interactions with adjacent territory holders as well as loons from other lakes)
and therefore must remain on the surface for longer periods to successfully defend their
territory. Additionally, larger lakes potentially support a greater number and diversity of
nesting locations and habitat variation, which may be preferred by loons. The information
for the lake attribute fields was gathered from the Environmental Protection Agency
Wildlife Database Search Engine (http://oaspub.epa.gov/aed/wildlife.search) and the
USGS National Water Information System (http://nwis.waterdata.usgs.gov). We also
accounted for spatial variation at the regional scale by including latitude and longitude
(decimal degrees).
We included three characteristics of the social environment previously found to
influence loon behavior. First, we characterized each loon territory as one of three types.
Loons breed on multiple lake territories (where their territory is comprised of many
77
lakes), whole lake territories (where there is one pair of loons on one lake), and partial
lake territories (where more than one pair is on a given lake; Evers et al. 2010). All of
the territories in this study were either whole lake or partial lake types. We predicted that
loons on single-lake territories would encounter lower energetic demands as they do not
defend shared territory boundaries with other pairs, nor do they need to fly to other lakes
to meet their nutritional requirements, as they do on multi-lake territories (Piper et al.
1997b). Second, we included both the number of pairs on the lake and the interaction
between number of pairs and lake surface area to account for potential increases in
territorial interactions with bird density on various lake sizes.
We accounted for temporal variation in the triglyceride concentration by
including six potential variables. We included the sampling year to account for year-to-
year variation and julian date to control for seasonal variation. To account for the
potential impacts of the local weather prior to sampling on triglyceride levels, we
included the minimum and maximum daily temperature (to the nearest tenth of a degree
C) on the day prior to the night of capture and the two, two-way interactions between
temperature (minimum and maximum) and year to account for variation in this
relationship among years. Weather data were collected from the National Oceanic and
Atmospheric Administration’s National Climatic Data Center
(http://www.ncdc.noaa.gov/) from the closest land-based weather station with historical
data for the day in question.
Lastly we included four variables that controlled for intrinsic individual
differences. We included sex to detect possible differences between male and female
triglyceride levels (e.g. energetic demands of incubation; (Williams 1996, Thomson et al.
78
1998, Evers et al. 2010; or territory defense; Piper et al. 2008) and the number of chicks
produced (1 or 2) to account for potential differences in the energetic demands of
different brood sizes (Deerenberg et al. 1995, Lozano and Lemon 1998). As body mass
is an important variable for interpreting metabolite results (Jenni-Eiermann and Jenni
1994), we included size-corrected body mass (weight/tarsus) in our model. Lastly, we
included free glycerol concentration as a candidate predictor, because the relationship
that circulating triglycerides have with free glycerol can help explain the stage of fat
acquisition or depletion that an organism is experiencing (Jenni-Eiermann and Jenni
1994, Guglielmo et al. 2005).
Candidate models were all repeated-measures, mixed-effects models (LMM)
using maximum likelihood (“ML”; package ‘nlme’ in R 2.14.1; Jose Pinheiro et al. 2011)
with the lake of capture was a random effect. Model selection was conducted using
stepwise AIC model selection (“stepAIC”, package ‘MASS’ in R 2.14.1; (Venables and
Ripley 2009), using backward selection. Optimal variance and correlation structures for
the models were selected using Akaike’s Information Criterion (AIC) values (Pinheiro
and Bates 2000, Zuur et al. 2009), and models with ∆AIC < 2.0 were considered
equivalent.
For our analysis of behavioral data, we used Microsoft Office Excel (2007) to
determine the mean and standard error of our observations.
3.4 RESULTS
3.4.1. Establishing Baseline Metabolic Measures
Loons of differing mass were largely homogenous in energetic condition which
allowed us to define the values for energetic regulation of known breeding individuals.
79
Using plasma samples (n=96), we established baseline levels of free glycerol (x SD =
0.157 ± 0.080 mmol/L) and triglyceride levels (0.875 ± 0.376) for brooding loons (Table
6).
Table 6. Descriptive Statistics for Triglyceride and Free Glycerol Levels. Descriptive
statistics for the average triglyceride and glycerol levels of common loons (Gavia immer) with
chicks in areas of ME, NH, MA, MT, WA during the summer of 2010 and 2011.
n mean SD
Free Glycerol
(mmol/L)
105 0.157 0.080
Triglycerides
(mmol/L)
105 0.875 0.376
Our values were within the range of values found in (or extrapolated by) previous
studies (see Chapter 3 of thesis). Our mean free glycerol levels are qualified as “low”
based on previous measures of glycerol, indicating birds in a fasting stage (Jenni-
Eiermann and Jenni 1998). Our mean triglyceride levels are also categorized as “low”
(as we had predicted based on our nocturnal capture and breeding stage), but exhibited
more variation than our glycerol values (51% vs. 43% of the mean).
3.4.2. Triglyceride Levels In Breeding Loons
Our model selection took eleven steps to identify three, equivalent models (∆AIC
< 2; Table 7) of triglyceride variation that included nine variables (in addition to the
random effect of the lake of capture) (cumulative Akaike wi, > 0.8). The parameter
estimates for number of chicks, latitude, and the interaction between year and maximum
temperature during the day preceding capture were all significantly different from zero.
The number of chicks was negatively related to blood triglycerides in the adults and the
latitude of capture was positively related to triglyceride concentrations. There was a
80
significant negative relationship between the highest maximum temperatures and
triglycerides in 2011 but not in 2010. Akaike weights for the top three models (out of
eleven potential models) were 32.4%, 26.2%, and 24.1% respectively (Table 7). Number
of chicks, year, territory type, latitude, free glycerol levels, maximum temperature, and
the interaction of year and maximum temperature were found in all eleven models. We
employed model averaging of the eleven models to create parameter estimates as a way
to include model selection uncertainty into these values (Table 5). Model selection
indicated that the inclusion of julian date, year, free glycerol levels, and longitude
improved model support, but the 95% confidence intervals of these parameter estimates
included zero, making it difficult to gauge their strength (Table 8).
Size-corrected body mass had no influence on triglyceride levels (was not a
covariate in our top three models). A post-hoc analysis showed no relationship between
triglyceride levels and the residuals of size-corrected body mass (r2= 0.003).
81
Table 7. Triglyceride Probability Models. Random effects model selection results for factors
affecting the level of triglycerides in plasma samples collected from common loons in ME, NH,
MA, MT, and WA, United States (2010 & 2011). The first three models performed equivalently
(∆AIC < 2), and all models contain the variables in the top-listed model in addition to those in
each row. ln(L) is the value of the maximized log-likelihood function, AIC is Akaike's
information criterion, ∆AIC is the scaled AIC relative to the top model, k is the number of model
parameters, and wi is the Akaike weight, which provides a relative measure of support for a given
model. Models were fit with maximum likelihood (ML). Variable abbreviations are as follows:
chicks (number of chicks a pair was caring for), year (capture year), fg (free glycerol levels in
mmol/L), terrType (either partial lake or whole lake territory), Lat (latitude), Long (longitude),
TMAX (maximum temperature during the day preceding the night of capture), TMIN (minimum
temperature during the day preceding the night of capture), year:TMAX (the interaction of year
and maximum temperature during the day preceding the night of capture), year:TMIN (the
interaction of the year and minimum temperature during the day preceding the night of capture),
Julian (Julian calendar date of capture day), sex (sex of individual), numPairs (the number of
pairs holding a territory on the lake of the captured individual), lakesurf (lake surface area in
hectares), numPairs:lakesurf (the interaction of lake surface area with the number of pairs holding
a territory on the lake), watershed (the area of the watershed in hectares), SZ_COR_BM (size-
corrected body mass; mass divided by tarsus width).
model ln(L) AIC ∆AIC k wi
m1 = chicks + year + fg+ terrType + Lat + TMAX +
year:TMAX -50.790 121.58 0 7 0.324
m1 + Long -50.004 122.01 0.43 8 0.262
m1 + Long + julian -49.087 122.17 0.59 9 0.241
m1 + TMIN + Julian + Long -49.081
124.16 2.58 10 0.089
m1 + TMIN + julian + Long + year:TMIN -48.683
125.37 3.79 11 0.049
m1 + TMIN + julian + Long + sex +
year:TMIN -48.566 127.13 5.55 12 0.020
m1 + TMIN + numPairs + sex+ julian + Long +
year:TMIN -48.492
128.98 7.4 13 0.008
m1 + TMIN + numPairs + lakesurf+ sex +
julian + Long + year:TMIN -48.446 130.89 9.31 14 0.003
m1 + TMIN + numPairs + lakesurf + sex +
julian + Long + year:TMIN +
numPairs:lakesurf
-48.400 132.80 10.78 15 0.001
m1 + TMIN + numPairs + lakesurf + sex +
julian + Long + year:TMIN +
numPairs:lakesurf + watershed
-48.395 134.79 11.22 16 <0.001
m1 + TMIN + numPairs + watershed +
lakesurf + sex + julian + Long + year:TMIN +
numPairs:lakesurf + SZ_COR_BM -48.390 136.78 15.2 17 <0.001
82
Table 8. Triglyceride Model-Averaged Parameter Estimates. Parameter estimates from
original 11 models and model-averaged parameter estimates with unconditional 95% confidence
intervals for variables in the top three models for triglyceride levels in common loons (Gavia
immer) in ME, NH, MA, MT, and WA, 2010 and 2011. Parameters considered include the
number of chicks (0 or 1), year (2010 or 2011), territory type (single lake or shared lake), capture
date (Julian), latitude and longitude of capture lake, maximum temperature on day of capture, and
the interaction between year and maximum temperature. Values in the parentheses following
each variable name indicate the reference level for binomial variables. Model-averaged
parameters whose estimates are significantly different from zero are in bold text.
Variable fixed effects
coefficient 95% CI
Chicks -0.209 -0.3935, -0.0264
Year (2011) 1.202 -0.2816, 2.6844
Territory Type
(partial-lake territory) -0.221 -0.4848, 0.0433
Julian -0.005 -0.0128, 0.0024
free glycerol 0.929 -0.3824, 2.2413
Degrees North Latitude 0.125 0.0467, 0.3826
Degree East Longitude 0.012 -0.0052, 0.0293
Maximum Temperature on Previous Day 0.001 -0.0026, 0.0040
Year (2011) x Maximum Temp -0.006 -0.0122, -0.0001
3.4.3. Behavior
Among 1228 observations of 113 loon territories, we recorded 56 incidences of
territorial intrusion on 32 territories. Only a single territorial pair ever exhibited both
escalated and un-escalated (i.e. active and passive) territorial behaviors. For the 22 loon
pairs for which both behavioral data and metabolite samples were collected, eight showed
territorial behaviors, and loons that escalated interactions had lower plasma triglyceride
concentrations than those that did not (Table 9), although the sample size is quite small.
83
Table 9. Descriptive statistics for the average triglyceride levels of common loons (Gavia immer)
with intruding loons on their territory that either did not respond aggressively (passive) or
responded with some degree of aggression (excalating) in ME, NH, MA during the summer of
2010 and 2011.
x n SE
passive 0.957 3 0.1365
escalating 0.784 5 0.0225
3.5 DISCUSSION
Our triglyceride levels suggest the loons we captured were in a fasting stage
(lower overall triglyceride values, i.e. <2.0 mmol/L), which is corroborated by the timing
of our sampling (nocturnal sampling in a diurnal foraging species). Within our range of
values, loons with lower triglycerides have higher energetic demand as they are utilizing
and mobilizing lipids more rapidly than the birds with higher plasma triglyceride levels
(Jenni-Eiermann and Jenni 1998, Nagy et al. 1999, Price et al. 2008). Loons that
escalated interactions had lower blood triglyceride concentrations pointing to an
increased cost in escalating interactions with intruding males, as suggested by previous
studies (Piper et al. 2000)
Our models suggested that energetic demand in breeding loons is influenced by
factors on various scales. Our top-ranked models suggest that energetic condition as
assessed by triglycerides is a function of a loon’s individual breeding effort (the number
of chicks produced), region (latitude, longitude), and weather conditions (maximum
temperature and the interaction of year and temperature), controlling for the social
environment (territory type), other metabolites (glycerol), and the time of sampling
(julian capture date, year).
84
The number of chicks a loon cared for was negatively related to triglyceride
values (when controlling for other variables in the model), which suggests that within our
range of values, loons with one chick are less energetically taxed than those with two.
Similar to our results, Deerenberg et al. (1995) found an increase in daily energy
expenditure in European kestrels (Falco tinnunculus) with increased clutch size. More
indirectly, decreases in mass are frequently reported during breeding in birds (for a
review see Moreno 1989), suggesting increases in energetic demand are related to
offspring production. This increase in energetic demand is likely a function of the
required rate of provisioning, which is a function of brood size (Jodice et al. 2006,
Gingras and Paszkowski 2006, Enstipp et al. 2007). Loon chicks are completely
dependent on their parents for food during the first several days of life. By three weeks
of age, they can chase fish for upwards of 30 meters, however their capture rate is still
poor (3%), and parents supply chicks with most of their food until nearly eleven weeks of
age (Gingras and Paszkowski 2006, Evers et al. 2010). Yellow warblers (Dendroica
petechia) increase both their provisioning rate and energetic demand when clutch size is
experimentally increased (Lozano and Lemon 1998), and a similar mechanism is likely at
play in our system. Our results, therefore, suggest that a future drop in the number of
loons brooding two versus one chick may signal an underlying physiological inability to
care for a larger brood, which could signify larger shifts toward population inviability.
What is unknown in our study species is how previous breeding experience impacts
current breeding condition at a given brood size as has been shown in common terns
(Sterna hirundo; Bauch et al. 2010).
85
Latitude was positively associated with triglyceride levels in our sample of loons,
indicating that at more northern sites, loons face decreased energetic demand. Our
latitude values ranged from 43.2 ˚ N to 48.8 ˚N. Basal metabolic rate has been shown to
correlate positively within vertebrate phylogenies with latitude (Weathers 1979, 1997,
Weathers and Greene 1998). Our results are contrary to this larger pattern, but it is
difficult to determine whether this is due to the relatively small latitudinal range over
which we conducted the study or whether this is indeed a physiological symptom of a
range edge. Collecting metabolites over a wider range of latitudes and along northern as
well as southern range edges would help illuminate the generality of this trend.
Our results indicated that triglyceride levels were negatively related to maximum
daily temperature on the day of capture on a given year (Fig. 15), suggesting that higher
maximum temperature produced lower energetic demand in 2011 (while controlling for
the other variables in the model, including latitude). Energy expenditure has been
reported to decrease with ambient temperature in other bird species as well (Claudia
Rauter and Reyer 2000). The strong relationship between metabolic demand and
temperature was present only in 2011. We assume that the metabolic demand of loons is
even more tightly tied to water temperature, since heat loss by endotherms in water is
higher than in air at the same temperature due to the thermal conductivity of water
(Jenssen et al. 1988). While water birds have mechanisms for increased thermoregulation
owing to their aquatic lifestyle (e.g. plumage insulation, peripheral vasoconstriction),
minimal thermal “wet” conductance in some waterbirds can be as much as 5 % greater
than conductance due to air (Jenssen et al. 1988). Unfortunately, we did not record
surface water temperatures during capture. As mean global temperature increases (IPCC,
86
2007), however, loons may be more energetically taxed due to higher maximum
temperatures during the chick rearing stage. Additionally, incubation is a
thermoregulatory challenge for loons and other birds under warm temperatures (Eppley
1996). The loons within our study area regularly pant during incubation (which suggests
they are above their thermal neutral zone), suggesting that higher temperatures may
impact the overall cost of breeding more dramatically outside of the chick rearing stage.
Higher temperatures over the long term will also likely influence prey species
abundances and biomass (Ficke et al. 2007). Additionally, higher surface water
temperatures will increase the strength and duration of lake stratification (Peeters et al.
2007), altering lake communities and food-web dynamics (Ficke et al. 2007). While
loons are adapted to surviving on various different prey items (Evers et al. 2010), large-
scale shifts in prey composition could potentially push loons out of once suitable
breeding lakes regardless of any decrease in energetic expenditure on thermal regulation.
87
Figure 15. Temperature/Year Interaction Plot. An all-effects plot examining the relationship
of year and temperature on triglyceride levels in common loons (Gavia immer), May- August,
2010-2011. Maximum temperature values on capture days ranged from 16.6-32.2 degrees C in
2010 and 17.9-35.0 degrees C in 2011. The relationship with the highest maximum temperatures
is shown in the upper plot and lowest maximum temperature in the lower plot.
3.6 SUMMARY
Breeding birds must optimize the allocation of energy into self-maintenance and
parental care (Martins and J 1993, Pettifor 1993) in the face of varying climatic
conditions and food resources (Claudia Rauter and Reyer 2000). Blood triglyceride
concentrations suggest that the number of chicks, breeding latitude, and daily temperature
are important predictors of loon energy expenditure (controlling for year, territory type,
and calendar date, free glycerol levels, size-corrected body mass, and longitude).
Specifically we found that: 1) loons with two chicks expend more energy than those with
one, 2) loons near the southern range edge expend more energy to produce a given brood
size than those nearer the range center and, 3) birds breeding in warmer temperatures
expend more energy than those in cooler temperatures (controlling for year, territory
88
type, and calendar date, free glycerol levels, size-corrected body mass, and longitude).
Without taking further measures of physiology, biologists can monitor maximum
temperature across years, and chick number (and latitude, if examining larger areas) and
identify areas of vulnerability before reproduction halts entirely. Of critical importance
as climate changes is monitoring changes in maximum temperature. While our findings
suggest that warmer temperatures may reduce the cost of thermoregulation for breeding
loons, our concern is in regards to the ways that temperature is predicted to negatively
impact hydrologic systems. Together these results suggest that common loons may be
sensitive to increases in temperature along their southern range boundary. Previous
investigations of southern range determinants using either brood counts or bird presence
did not detect similar patterns, which suggests that physiological measures may be more
sensitive indicators of range contractions than more traditional bioclimatic envelope or
demographic approaches.
89
BIBLIOGRAPHY
Alonso-Alvarez, C., and M. Ferrer. 2001. A biochemical study of fasting, subfeeding, and
recovery processes in yellow-legged gulls. Physiological and biochemical zoology
74:703–13.
Alvo, R., D. Hussell, and M. Berrill. 1988. The breeding success of common loons (
Gavia immer ) in relation to alkalinity and other lake characteristics in Ontario.
Canadian Journal of Zoology 66:746–752.
Armstrong, R. L., and M. J. Brodzik. 2006. Northern Hemisphere Snow Extent Derived
from Microwave and Optical Satellite Data. Pages 1–7. . Boulder, Colorado.
Badzinski, S. S., and S. T. A. Timmermans. 2006. Factors influencing productivity of
common loons (Gavia immer) breeding on circumneutral lakes in Nova Scotia,
Canada. Hydrobiologia 567:215–226.
Barbet-Massin, M., B. a Walther, W. Thuiller, C. Rahbek, and F. Jiguet. 2009. Potential
impacts of climate change on the winter distribution of Afro-Palaearctic migrant
passerines. Biology letters 5:248–51.
Barnett, T. P., J. C. Adam, and D. P. Lettenmaier. 2005. Potential impacts of a warming
climate on water availability in snow-dominated regions. Nature 438:303–9.
Barr, J. F. 1973. Feeding biology of the Common Loon (Gavia immer) in oligotrophic
lakes of the precambrian shield. . Univ. of Guelph, Ontario.
Bauch, C., S. Kreutzer, and P. H. Becker. 2010. Breeding experience affects condition:
blood metabolite levels over the course of incubation in a seabird. Journal of
comparative physiology. B, Biochemical, systemic, and environmental physiology
180:835–45.
Beale, C. M., J. J. Lennon, and A. Gimona. 2008. Opening the climate envelope reveals
no macroscale associations with climate in European birds. Proceedings of the
National Academy of Sciences of the United States of America 105:14908–12.
Blair, R. B. 1992. Lake Features, Water Quality and the Summer Distribution of
Common Loons in New Hampshire. Distribution 63:1–9.
Blenckner, T. 2005. A conceptual model of climate-related effects on lake ecosystems.
Hydrobiologia 533:1–14.
Breiman, L. 2001. Random Forests. Machine Learning 45:5–32.
Breiman, L., J. Friedman, C. J. Stone, and R. A. Olshen. 1984. Classification and
Regression Trees. . Chapman & Hall/CRC.
90
Brown, J. L., S. H. Li, and N. Bhagabati. 1999. Long-term trend toward earlier breeding
in an American bird: a response to global warming? Proceedings of the National
Academy of Sciences of the United States of America 96:5565–69.
Byrd, A. J., B. J. Olsen, D. C. Evers, and R. L. Holberton. 2012. Common Loon (Gavia
immer) Biogeography & Reproductive Success in an Era of Climate Change.
Carey, C. 1996. Avian Energetics and Nutritional Ecology. Pages 324–374. . Chapman &
Hall/CRC, New York, New York, USA.
Chase, M. K., N. Nur, and G. R. Geupel. 2005. Effects of Weather and Population
Density on Reproductive Success and Population Dynamics in a Song Sparrow
(Melospiza melodia ) Population : A Long-Term Study. The Auk 122:571–592.
Claudia Rauter, and H. Reyer. 2000. Thermal and energetic consequences of nest location
and breeding times in Water Pipits (Anthus spinoletta). Journal of Ornithology
141:391–407.
Corney, P. M., M. G. Le Duc, S. M. Smart, K. J. Kirby, R. G. H. Bunce, and R. H. Marrs.
2006. Relationships between the species composition of forest field-layer vegetation
and environmental drivers, assessed using a national scale survey. Journal of
Ecology 94:383–401.
Crawley, M. 2007. The R Book. . Wiley, Chichester, England; Hoboken, N.J.
Cutler, D. R., T. C. Edwards, K. H. Beard, A. Cutler, K. T. Hess, J. Gibson, and J. J.
Lawler. 2007. Random forests for classification in ecology. Ecology 88:2783–92.
Dale, V. 1997. The Relationship Between Land-Use Change and Climate Change.
Ecological Applications 7:753–769.
Davis, A. J., L. S. Jenkinson, J. H. Lawton, B. Shorrocks, and S. Wood. 1998a. Making
mistakes when predicting shifts in species range in response to global warming.
Nature 391:783–786.
Davis, A. J., J. H. Lawton, B. Shorrocks, and L. S. Jenkinson. 1998b. Individualistic
species responses invalidate simple physiological models of community dynamics
under global environmental change. Journal of Animal Ecology 67:600–612.
Deerenberg, C., I. Pen, C. Dijkstra, B.-J. Arkies, G. H. Visser, and S. Daan. 1995.
Parental energy expenditure in relation to manipulated brood size in the european
kestrel. Zoology, Analysis of Complex Systems (ZACS) 99:38–47.
Desorbo, C. R., K. M. Taylor, D. E. Kramar, J. Fair, J. H. Cooley, D. C. Evers, W.
Hanson, H. S. Vogel, and J. L. Atwood. 2007. Reproductive Advantages for
Common Loons Using Rafts. Journal of Wildlife Management 71:1206–1213.
91
De’ath, G., and K. Fabricius. 2000. Classification and regression trees: a powerful yet
simple technique for ecological data analysis. Ecology 81:3178–3192.
Doligez, B., E. Danchin, J. Clobert, and L. Gustafsson. 1999. The use of conspecific
reproductive success for breeding habitat selection in a non-colonial, hole-nesting
species, the collared flycatcher. Journal of Animal Ecology 68:1193–1206.
Drilling, N. E., and C. F. Thompson. 1988. Natal and breeding dispersal in house wrens
(troglodytes-aedon). Auk 105:480–491.
Driscoll, C. T., K. M. Driscoll, K. M. Roy, and M. J. Mitchell. 2003. Chemical response
of lakes in the Adirondack Region of New York to declines in acidic deposition.
Environmental science & technology 37:2036–42.
Dyke, A. S., and V. K. Prest. 1987. Late Wisconsinan and Holocene History of the
Laurentide Ice Sheet. Geographie physique et Quaternaire 41:2.
Enstipp, M. R., D. R. Jones, S.-H. Lorentsen, and D. Grémillet. 2007. Energetic costs of
diving and prey-capture capabilities in cormorants and shags (Phalacrocoracidae)
underline their unique adaptation to the aquatic environment. Journal of Ornithology
148:593–600.
Eppley, Z. A. 1996. Society for Experimental Biology Seminar Series 59. Pages 313–346
Development of Thermoregulation in Birds: physiology, interspecific variation and
adaptation to climate.
Evers, D. C. 1992. A replicable capture method for adult and juvenile common loons on
their nesting lakes. Proc. From the 1992 conference on the loon and its ecosystem:
Status, management, and environmental concerns.:214–220.
Evers, D. C. 2001a. Common Loon Population Studies: Continental Mercury Patterns
and Breeding Territory Philopatry. . Univ. of MN, St. Paul, MN.
Evers, D. C. 2001b. Common Loon Population Studies: Continental Mercury Patterns
and Breeding Territory Philopatry.
Evers, D. C., J. D. Paruk, J. W. McIntyre, and J. F. Barr. 2010. Common Loon (Gavia
immer).
Evers, D. C., L. J. Savoy, C. R. DeSorbo, D. E. Yates, W. Hanson, K. M. Taylor, L. S.
Siegel, J. H. Cooley, M. S. Bank, A. Major, K. Munney, B. F. Mower, H. S. Vogel,
N. Schoch, M. Pokras, M. W. Goodale, and J. Fair. 2008. Adverse effects from
environmental mercury loads on breeding common loons. Ecotoxicology (London,
England) 17:69–81.
92
Ewert, D. 1982. Spring migration of Common Loons at Whitefish Point, Michigan. Jack-
Pine Warbler 60:134–143.
Ficke, A. D., C. a. Myrick, and L. J. Hansen. 2007. Potential impacts of global climate
change on freshwater fisheries. Pages 581–613 Reviews in Fish Biology and
Fisheries.
Gaston, K. J. 2003. The structure and dynamics of geographic ranges. Oxford University
Press.
Gingras, B., and C. a. Paszkowski. 2006. Feeding behavior and modeled energetic intake
of common loon (Gavia immer) adults and chicks on small lakes with and without
fish. Hydrobiologia 567:247–261.
Gostomski, T. J., and D. C. Evers. 1998. Time-activity budget for Common Loons, Gavia
immer, nesting on Lake Superior. Canadian Field-Naturalist 112:191–197.
Grear, J. S., M. W. Meyer, J. H. Cooley, A. Kuhn, H. Walter, M. G. Mitro, H. S. Vogel,
K. M. Taylor, K. P. Kenow, S. M. Craig, D. E. Nacci, A. Press, and S. Grear. 2009.
Population Growth and Demography the Northern United States of Common Loons
in. The Journal of Wildlife Management 73:1108–1115.
Greenwood, P. J., and P. H. Harvey. 1982. The natal and breeding dispersal of birds.
Annual Review of Ecology and Systematics 13:1–21.
Guglielmo, C. G., D. J. Cerasale, and C. Eldermire. 2005. A field validation of plasma
metabolite profiling to assess refueling performance of migratory birds.
Physiological and biochemical zoology 78:116–25.
Hampe, A. 2004. Bioclimate envelope models : what they detect and what they hide.
Global Ecology & Biogeography 13:469–476.
Heikkinen, R. K., M. Luoto, M. B. Araujo, R. Virkkala, W. Thuiller, and M. T. Sykes.
2006. Methods and uncertainties in bioclimatic envelope modelling under climate
change. Progress in Physical Geography 30:751–777.
Hemberger, M., D. Euler, and J. Barr. 1983. The impact of cottage development on
common loon reproductive success in central Ontario. Wilson Bulletin 95:431–439.
Herlihy, A. T., J. L. Stoddard, and C. B. Johnson. 1998. The relationship between stream
chemistry and watershed land cover data in the mid-atlantic region, U.S. Water, Air,
and Soil Pollution 105:377–386.
Hijmans, R. J., and C. H. Graham. 2006. The ability of climate envelope models to
predict the effect of climate change on species distributions. Global Change Biology
12:2272–2281.
93
Hitch, A. T., and P. L. Leberg. 2007. Breeding distributions of north American bird
species moving north as a result of climate change. Conservation biology : the
journal of the Society for Conservation Biology 21:534–9.
Hodkinson, I. D. 1999. Species response to global environmental change or why
ecophysiological models are important: a reply to Davis et al. Journal of Animal
Ecology 68:1259–1262.
Holberton, R. L. 2008. Matericals and basic methods for field blood collection, storage,
etc. . Orono, ME.
Holt, R. D., and T. H. Keitt. 2005. Species ’ borders : a unifying theme in ecology. Oikos
108:3–6.
Holt, R. D., T. H. Keitt, M. A. Lewis, B. A. Maurer, and M. L. Taper. 2005. Theoretical
models of species ’ borders : single species approaches. Oikos 108:18–27.
IPCC. 2007. Contribution of Working Group I to the Fourth Assessment Report of the
Intergovernmental Panel on Climate Change. Page 996 (S. D. Solomon, M. Qin, M.
Manning, Z. Chen, M. Marquiz, K. B. Averyt, M. Tignor, and H. L. Miller, Eds.). .
Cambridge University Press, United Kingdom and New York.
Jacobson, G. L., I. J. Fernandez, P. A. Mayewski, and C. V. Schmitt. 2009. Maine’s
Climate Future: An Initial Assessment. . Orono.
Jenni, L., and R. Schwilch. 2001. Plasma metabolite levels indicate change in body mass
in reed warblers Acrocephalus scirpaceus. Avian Science 1:55–65.
Jenni-Eiermann, S., and L. Jenni. 1994. Plasma Metabolite Levels Predict Individual
Body-Mass Changes in a Small Long-Distance Migrant , the Garden Warbler
Author. The Auk 111:888–899.
Jenni-Eiermann, S., and L. Jenni. 1997. Diurnal Variation of Metabolic Responses to
Short-Term Fasting in Passerine Birds during the Postbreeding , Molting and
Migratory Period. The Condor 99:113–122.
Jenni-Eiermann, S., and L. Jenni. 1998. What can plasma metabolites tell us about the
metabolism, physiological state and condition of individual birds? An overview.
Biological Conservation Fauna 102:312–319.
Jenssen, B. M., M. Ekker, and C. Bech. 1988. Thermoregulation in winter-acclimatized
common eiders (Somateria mollissima) in air and water. Canadian Journal of
Zoology 67:669–673.
94
Jodice, P. G. R., D. D. Roby, R. M. Suryan, D. B. Irons, K. R. Turco, E. D. Brown, J. F.
Thedinga, and G. H. Visser. 2006. Increased energy expenditure by a seabird in
response to higher food abundance. Marine Ecology Progress Series 306:283–293.
Jose Pinheiro, D. Bates, S. DebRoy, and D. Sarkar. 2011. nlme: Linear and Nonlinear
Mixed Effects Models. R package version 3.1-106.
Karasov, W. H., K. P. Kenow, M. W. Meyer, and F. Fournier. 2007. Bioenergetic and
pharmacokinetic model for exposure of common loon (Gavia immer) chicks to
methylmercury. Environmental toxicology and chemistry / SETAC 26:677–85.
Kaushal, S. S., P. M. Groffman, G. E. Likens, K. T. Belt, W. P. Stack, V. R. Kelly, L. E.
Band, and G. T. Fisher. 2005. Increased salinization of fresh water in the
northeastern United States. Proceedings of the National Academy of Sciences of the
United States of America 102:13517–20.
Keating, J. 2004. Deicing Salt: Still on the Table. Stormwater.
Kenow, K. P. K., D. A. Dams, N. Schoch, D. Evers, W. Hanson, D. Yates, L. Savoy, T.
Fox, A. Major, R. Kratt, J. Ozard, R. O. K. Ratt, and J. O. H. N. O. Zard. 2009.
Migration Patterns and Wintering Range of Common Loons Breeding in the
Northeastern United States. Waterbirds 32:234–247.
Kenow, K. P., M. W. Myer, R. K. Hines, and W. I. H. K. Karasov. 2007. Distribution and
accumulation of mercury in tissues of captive-reared common loon ( gavia immer )
chicks. Environmental toxicology and chemistry / SETAC 26:1047–1055.
Kingston, J. D., A. L. Deino, R. K. Edgar, and A. Hill. 2007. Astronomically forced
climate change in the Kenyan Rift Valley. Journal of Human Evolution 53:487–503.
Kuhn, A., J. Copeland, J. Cooley, H. Vogel, K. Taylor, D. Nacci, and P. August. 2011.
Modeling Habitat Associations for the Common Loon ( Gavia immer ) at Multiple
Scales in Northeastern North America. Avian Conservation and Ecology 6.
Lathrop, R. C., S. R. Carpenter, and D. M. Robertson. 1999. water clarity responses to
phosphorus , Daphnia grazing , and internal in Lake Mendota mixing. Limnology
and Oceanography 44:137–146.
Laudon, H., P. J. Dillon, M. C. Eimers, R. G. Semkin, and D. S. Jeffries. 2004. Climate-
induced episodic acidification of streams in central ontario. Environmental science
& technology 38:6009–15.
Liaw, A., and M. Wiener. 2012. Package “randomForest”. . CRAN.
95
Lozano, G. A., and R. E. Lemon. 1998. Parental-care responses by yellow warblers
(Dendroica petechia) to simultaneous manipulations of food abundance and brood
size. Canadian Journal of Zoology 76:916–924.
Mager, J. N., and C. Walcott. 2007. Structural and contextual characteristics of territorial
“yodels” given by male Common Loons (Gavia immer) in northern Wisconsin.
Passenger Pigeon 69:327–337.
Mager, J. N., C. Walcott, and W. H. Piper. 2007. Male Common Loons, Gavia immer,
communicate body mass and condition through dominant frequencies of territorial
yodels. Animal behaviour 73:683–690.
Magnuson, J. J., K. E. Webster, R. A. Assel, C. J. Bowser, P. J. Dillon, J. G. Eaton, H. E.
Evans, E. J. Fee, R. I. Hall, L. R. Mortsch, D. W. Schindler, and F. H. Quinn. 1997.
Potential effects of climate changes on aquatic systems : laurentian great lakes and
precambrian shield region. Hydrological Processes 11:825–871.
Martinez-Morales, M. A., I. Zuria, L. Chapa-Vargas, I. MacGregor-Fors, R. Ortega-
Alvarez, E. Romero-Aguila, and P. Carbo. 2010. Current distribution and predicted
geographic expansion of the Rufous-backed Robin in Mexico: a fading endemism?
Diversity and Distributions 16:786–797.
Martins, T., and W. J. 1993. Cost of reproduction and allocation of food between parent
and young in the swift (Apus apus). Behavioral Ecology 4:213±223.
Matthews, S. N. 2004. Atlas of climate change effects in 150 bird species of the Eastern
United States. . US Department of Agriculture, Forest Service, Northeastern
Research Station.
McIntyre, J. W. 1994. Loons in freshwater lakes. Hydrobiologia 279-280:393–413.
McIntyre, J. W. ., and J. F. . Barr. 1983. Pre-Migratory Behavior of Common Loons on
the Autumn Staging Grounds. Wilson Ornithological Society 95:121–125.
McNicol, D. K., M. L. Mallory, and H. S. Vogel. 1995. Using volunteers to monitor the
effects of acid precipitation on common loon (Gavia immer) reproduction in canada:
the canadian lakes loon survey. Water, Air, and Soil Pollution 85:463–468.
Moreno, J. 1989. Strategies of mass change in breeding birds. Biological Journal of the
Linnean Society:297–310.
Nagy, K. a, I. a Girard, and T. K. Brown. 1999. Energetics of free-ranging mammals,
reptiles, and birds. Annual review of nutrition 19:247–77.
96
Newman, S. H., J. F. Piatt, and J. White. 1997. Hematological and Plasma Biochemical
Reference Ranges of Alaskan Seabirds : Their Ecological Significance and Clinical
Importance. Colonial Waterbirds 20:492–504.
Nickerson, D. M., D. E. Facey, and G. D. Grossman. 1988. Estimating Physiological
Thresholds with Continuous Two-Phase Regression. Physiological Zoology 62:866–
887.
Nocera, J. J., and N. M. Burgess. 2002. Diving schedules of Common Loons in varying
environments. Canadian Journal of Zoology 80:1643–1648.
Olson, S. T., and W. H. Marshall. 1952. The Common Loon in Minnesota.
O’Connor, R. J., and T. L. Wagner. 2004. a Test of a Regression-Tree Model of Species
Distribution. The Auk 121:604.
Parker, K. E. 1988. Common Loon Reproduction and Chick Feeding on Acidified Lakes
in the Adirondack Park, New York. Canadian Journal of Zoology 66:804–810.
Paruk, J. D. 2009. Function of the common loon foot waggle. The Wilson Journal of
Ornithology 121:392–398.
Payne, R. B., and L. L. Payne. 1993. Breeding dispersal in Indigo Buntings:
circumstances and consequences for breeding success and population structure.
Condor 95:1–24.
Pearson, R. G., and T. P. Dawson. 2003. Predicting the impacts of climate change on the
distribution of species : are bioclimate envelope models useful ? Global Ecology &
Biogeography 12:361–371.
Peeters, F., D. Straile, A. Lorke, and D. Ollinger. 2007. Turbulent mixing and
phytoplankton spring bloom development in a deep lake. Limnology and
Oceanography 52:286–298.
Pettifor, R. A. 1993. Brood-manipulation experiments II. A cost of reproduction in blue
tits (Parus caeruleus)? Journal of Animal Ecology 62:145–159.
Pinheiro, J. C., and D. M. Bates. 2000. Mixed-effects models in S and S-PLUS. . New
York: Springer.
Piper, W. H., D. Evers, M. W. Meyer, K. B. Tischler, D. Kaplan, and R. C. Fleischer.
1997a. Genetic monogamy in the common loon (Gavia immer). Behavioral Ecology
and Sociobiology 41:25–31.
97
Piper, W. H., J. D. Paruk, D. C. Evers, M. W. Meyer, B. Keren, M. Klich, and J. J.
Hartigan. 1997b. Local Movements of Color-Marked Common Loons. The Journal
of Wildlife Management 61:1253–1261.
Piper, W. H., C. Walcott, J. N. Mager, and F. J. Spilker. 2008a. Nestsite selection by
male loons leads to sex-biased site familiarity. Journal of Animal Ecology 77:205–
210.
Piper, W. H., C. Walcott, J. N. Mager, and F. J. Spilker. 2008b. Fatal battles in common
loons: a preliminary analysis. Animal Behaviour 75:1109–1115.
Piper, W., K. Tischler, and M. Klich. 2000. Territory acquisition in loons: the importance
of take-over. Animal behaviour 59:385–394.
Poff, N. L. 2002. Ecological response to and management of increased flooding caused
by climate change. Philosophical transactions. Series A, Mathematical, physical, and
engineering sciences 360:1497–510.
Pokras, M. A., C. Hanley, and Z. Gordon. 1998. Liver mercury and methylmercury
concentrations in New England common loons (Gavia immer). Environmental
Toxicology and Chemistry 17:202–204.
Price, E. R., A. Krokfors, and C. G. Guglielmo. 2008. Selective mobilization of fatty
acids from adipose tissue in migratory birds. The Journal of experimental biology
211:29–34.
Pörtner, H. O. 2002. Climate variations and the physiological basis of temperature
dependent biogeography: systemic to molecular hierarchy of thermal tolerance in
animals. Comparative biochemistry and physiology. Part A, Molecular & integrative
physiology 132:739–61.
Quillfeldt, P., J. F. Masello, and E. Mostl. 2004. Blood chemistry in relation to nutrition
and ectoparasite load in Wilson’s storm-petrels Oceanites oceanicus. Polar Biology
27:168–176.
Ramakrishna, D. M., and T. Viraraghavan. 2005. Environmental Impact of Chemical
Deicers – A Review. Water, Air, and Soil Pollution 166:49–63.
Robinson, D. A., and R. R. Heim Jr. 2006. Trends and variability of snowfall and snow
cover across north america and eurasia. Part 2: what the data say. Pages 1–5. .
Asheville, NC.
Rodenhouse, N. L., S. N. Matthews, K. P. McFarland, J. D. Lambert, L. R. Iverson, a.
Prasad, T. S. Sillett, and R. T. Holmes. 2007. Potential effects of climate change on
birds of the Northeast. Mitigation and Adaptation Strategies for Global Change
13:517–540.
98
Rosenberry, D. O., P. A. Bukaveckas, D. C. Buso, G. E. Likens, A. M. Shapiro, and T. C.
Winter. 1999. MOVEMENT OF ROAD SALT TO A SMALL NEW HAMPSHIRE
LAKE. Water, Air, and Soil Pollution 109:179–206.
Rowley, I. 1983. Re-mating in birds. Pages 331–360 (P. Bateson, Ed.) Hydrobiologia. .
Cambridge University Press.
Ruggles, A. K. 1994. Habitat selection by loons in southcentral Alaska. Hydrobiologia
279-280:421–430.
Seavy, N. E., K. E. Dybala, and M. a. Snyder. 2008. Climate Models and Ornithology.
The Auk 125:1–10.
Sedgwick, J. A. 2004. Site fidelity, territory fidelity and natal philopatry in willow
flycatchers (Empidonax traillii). Auk 121:1103–1121.
Senar, J. C., and J. Pascual. 1997. Keel and Tarsus Length May Provide A Good
Predictor of Avian Body Size. Ardea 85:269–274.
Sidor, I. F., M. a Pokras, A. R. Major, R. H. Poppenga, K. M. Taylor, and R. M. Miconi.
2003. Mortality of common loons in New England, 1987 to 2000. Journal of wildlife
diseases 39:306–15.
Sonzogni, W. C., W. Richardson, P. Rodgers, and T. J. Monteith. 1983. Chloride
pollution of the Great Lakes. Journal of Water Pollution 55:513–521.
Stamps, J. A. 1987. Conspecifics as Cues to Territory Quality- A Preference of Juvenile
Lizards (Anolis-Aeneus) for Previously Used Territories. American Naturalist
129:629–642.
Stanley, T., and J. A. Royle. 2005. Estimating site occupancy and abundance using
indirect detection indices. Journal of Wildlife Management 69:3.
Switzer, P. V. 1997. Past reproductive success affects future habitat selection. Behavioral
Ecology and Sociobiology 40:307–312.
Therneau, T., B. Atkinson, and B. Ripley. 2012. Package “rpart.”
Thomas, C. D., E. J. Bodsworth, R. J. Wilson, a D. Simmons, Z. G. Davies, M. Musche,
and L. Conradt. 2001. Ecological and evolutionary processes at expanding range
margins. Nature 411:577–81.
Thomas, C. D., A. Cameron, R. E. Green, M. Bakkenes, L. J. Beaumont, Y. C.
Collingham, B. F. N. Erasmus, M. F. De Siqueira, A. Grainger, L. Hannah, L.
Hughes, B. Huntley, A. S. Van Jaarsveld, G. F. Midgley, L. Miles, M. a Ortega-
99
Huerta, a T. Peterson, O. L. Phillips, and S. E. Williams. 2004. Extinction risk from
climate change. Nature 427:145–8.
Thomas, C. D., and J. J. Lennon. 1999. Birds extend their ranges northwards. Nature
399:6505.
Thomson, D. L., P. Monaghan, and R. W. Furness. 1998. The demands of incubation and
avian clutch size. Biological Review 73:293^304.
Thuiller, W. 2003. BIOMOD - optimizing predictions of species distributions and
projecting potential future shifts under global change. Global Change Biology
9:1353–1362.
Totzke, U., M. Fenske, O. Hüppop, H. Raabe, and N. Schach. 1999. The influence of
fasting on blood and plasma composition of herring gulls (Larus argentatus).
Physiological and biochemical zoology 72:426–37.
Vallecillo, S., L. Brotons, and W. Thuiller. 2009. Dangers of predicting bird species
distributions in response to land-cover changes. Ecological applications : a
publication of the Ecological Society of America 19:538–49.
Venables, W. N., and B. D. Ripley. 2009. Modern Applied Statistics with S. Fourth
Edition. . Springer, New York.
Virkkala, R., R. K. Heikkinen, N. Leikola, and M. Luoto. 2008. Projected large-scale
range reductions of northern-boreal land bird species due to climate change.
Biological Conservation 141:1343–1353.
Waite, T. a, and D. Strickland. 2006. Climate change and the demographic demise of a
hoarding bird living on the edge. Proceedings. Biological sciences / The Royal
Society 273:2809–13.
Walsberg, G. E. 1983. Avian Biology. Pages 135–158.
Walther, G., E. Post, P. Convey, A. Menzel, C. Parmesank, T. J. C. Beebee, J. Fromentin,
O. H. I, and F. Bairlein. 2002. Ecological responses to recent climate change. Nature
416:389–395.
Watkinson, A. R., J. a. Gill, and M. Hulme. 2004. Flying in the face of climate change: a
review of climate change, past, present and future. Ibis 146:4–10.
Weathers, W. W. 1979. Climatic adaptation in avian standard metabolic rate. Oecologia
42:81–89.
Weathers, W. W. 1997. Energetics and thermoregulation by small passerines of the
humid, lowland tropics. Auk 114:341–353.
100
Weathers, W. W., and E. Greene. 1998. Thermoregulatory responses of bridled and
juniper titmice to high temperature. Condor 100:365–372.
Wetzel, R. 2005. Limnology. . Saunders, New York, New York, USA.
Williams, J. B. 1996. Energetics of avian incubation. Pages 375–415 Avian energetics
and nutritional ecology. . London: Chapman & Hall.
Williams, J. W., and S. T. Jackson. 2012. Novel ecological climates , No-Analog
Communities, and Ecological Surprises. Frontiers in Ecology and the Environment
5:475–482.
Wolfe, A. P., J. S. Baron, and R. J. Cornett. 2001. Anthropogenic nitrogen deposition
induces rapid ecological changes in alpine lakes of the Colorado Front Range ( USA
). Journal of Paleolimnology 25:1–7.
Zuur, A. F., E. N. Ieno, N. J. Walker, A. A. Saveliev, and G. M. Smith. 2009. Mixed
effects models and extensions in ecology with R. . New York: Springer.
101
BIOGRAPHY OF THE AUTHOR
Allison Jeanne Byrd was born in Flemington, NJ on June 28th
, 1979. She
graduated from Voorhees High School in 1997. She attended The University of Rhode
Island (Kingston, R.I.) and earned a B.S. in Wildlife Biology in 2001 after also
completing course work at an overseas course at the University of Bath, England. Allie
worked as a biological technician in the western United States (Washington, Wyoming,
and Montana), on Santa Cruz Island (California), and in the eastern United States (North
Carolina, Florida, New Jersey) studying birds and mammals. She also worked as a
biological technician abroad in Australia, Mexico, and Panama studying birds. She is a
candidate for the Master of Science degree in Ecology and Environmental Science from
the University of Maine in May, 2013.
Top Related